diff --git a/.gitattributes b/.gitattributes index a6344aac8c09253b3b630fb776ae94478aa0275b..52373fe24473b1aa44333d318f578ae6bf04b49b 100644 --- a/.gitattributes +++ b/.gitattributes @@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text *.zip filter=lfs diff=lfs merge=lfs -text *.zst filter=lfs diff=lfs merge=lfs -text *tfevents* filter=lfs diff=lfs merge=lfs -text +tokenizer.json filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000000000000000000000000000000000000..416106c533f62c763b0c1ec1b0653dff3dd4fcb1 --- /dev/null +++ b/README.md @@ -0,0 +1,38 @@ + +--- +license: llama3.1 +base_model: +- meta-llama/Llama-3.1-405B-Instruct +base_model_relation: quantized +tags: +- VPTQ +- Quantized +- Quantization +--- +**Disclaimer**: + +The model is reproduced based on the paper *VPTQ: Extreme Low-bit Vector Post-Training Quantization for Large Language Models* [github](https://github.com/microsoft/vptq) and [arXiv](https://arxiv.org/abs/2409.17066) + +The model itself is sourced from a community release. + +It is intended only for experimental purposes. + +Users are responsible for any consequences arising from the use of this model. + +**Note**: + +The PPL test results are for reference only and were collected using GPTQ testing script. + +```json +{ + "ctx_2048": { + "wikitext2": 1.9966317415237427 + }, + "ctx_4096": { + "wikitext2": 1.8661881685256958 + }, + "ctx_8192": { + "wikitext2": 1.8048142194747925 + } +} +``` diff --git a/config.json b/config.json new file mode 100644 index 0000000000000000000000000000000000000000..09fba166e8f00f9fec659be38d0cbca52f296c69 --- /dev/null +++ b/config.json @@ -0,0 +1,21210 @@ +{ + "_name_or_path": "meta-llama/Meta-Llama-3.1-405B-Instruct", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "bos_token_id": 128000, + "eos_token_id": [ + 128001, + 128008, + 128009 + ], + "head_dim": 128, + "hidden_act": "silu", + "hidden_size": 16384, + "initializer_range": 0.02, + "intermediate_size": 53248, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "num_attention_heads": 128, + "num_hidden_layers": 126, + "num_key_value_heads": 8, + "pretraining_tp": 1, + "quant_config": { + "model.layers.0.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.0.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.1.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.10.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.100.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.101.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.102.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.103.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.104.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.105.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.106.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.107.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.108.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.109.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.11.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.110.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.111.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.112.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.113.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.114.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.115.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.116.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.117.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.118.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.119.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.12.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.120.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.121.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.122.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.123.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.124.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.125.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.13.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.14.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.15.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.16.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.17.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.18.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.19.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.2.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.20.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.21.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.22.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.23.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.24.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.25.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.26.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.27.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.28.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.29.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.3.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.30.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.31.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.32.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.33.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.34.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.35.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.36.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.37.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.38.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.39.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.4.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.40.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.41.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.42.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.43.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.44.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.45.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.46.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.47.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.48.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.49.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.5.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.50.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.51.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.52.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.53.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.54.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.55.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.56.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.57.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.58.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.59.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.6.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.60.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.61.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.62.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.63.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.64.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.65.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.66.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.67.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.68.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.69.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.7.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.70.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.71.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.72.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.73.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.74.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.75.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.76.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.77.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.78.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.79.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.8.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.80.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.81.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.82.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.83.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.84.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.85.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.86.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.87.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.88.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.89.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.9.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.90.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.91.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.92.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.93.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.94.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.95.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.96.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.97.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.98.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.mlp.down_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 53248, + "in_features": 53248, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.mlp.gate_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.mlp.up_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 53248, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.self_attn.k_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.self_attn.o_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.self_attn.q_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 16384, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + }, + "model.layers.99.self_attn.v_proj": { + "bias": null, + "enable_norm": true, + "enable_perm": true, + "group_num": 1, + "group_size": 16384, + "in_features": 16384, + "indices_as_float": false, + "is_indice_packed": true, + "num_centroids": [ + -1, + 65536 + ], + "num_res_centroids": [ + -1, + 65536 + ], + "out_features": 1024, + "outlier_size": 0, + "vector_lens": [ + -1, + 8 + ] + } + }, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "torch_dtype": "bfloat16", + "transformers_version": "4.45.2", + "use_cache": true, + "vocab_size": 128256 +} diff --git a/generation_config.json b/generation_config.json new file mode 100644 index 0000000000000000000000000000000000000000..c76a0ab380db071b709078362d0a3187055ed821 --- /dev/null +++ b/generation_config.json @@ -0,0 +1,13 @@ +{ + "attn_implementation": "flash_attention_2", + "bos_token_id": 128000, + "do_sample": true, + "eos_token_id": [ + 128001, + 128008, + 128009 + ], + "temperature": 0.6, + "top_p": 0.9, + "transformers_version": "4.45.2" +} diff --git a/model-00001-of-00044.safetensors b/model-00001-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..cc6647b2c63f9319b5cbde3f0cf44aa8db7733cd --- /dev/null +++ b/model-00001-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e64bb430319d52e8f07a26f51cd1a8aaf2bcb7e4d4e9987d1c48a06a20c340d4 +size 4935192536 diff --git a/model-00002-of-00044.safetensors b/model-00002-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..02152bb444ca8c7d994cf44c902b692ce910cdfd --- /dev/null +++ b/model-00002-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1a9a03c637dde5e50226314bb7784cccfbc438e237245857a73fa5e56ce7e1d +size 4828487008 diff --git a/model-00003-of-00044.safetensors b/model-00003-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..12921352f51c66561945cfb04bb3af4dd0c20203 --- /dev/null +++ b/model-00003-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f48a81667167bc915f613a35426696992088d1f7bfe6ac3d440dc8bdd48f2366 +size 4828487008 diff --git a/model-00004-of-00044.safetensors b/model-00004-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..dc546ee35a417b92d116a1bfc7f466a740db3ee9 --- /dev/null +++ b/model-00004-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eec0e64f4aaf25c9771ff617bbef1d49739e210d214338325e818b783d39231f +size 4828487008 diff --git a/model-00005-of-00044.safetensors b/model-00005-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..71d759f3bab9d591b4b254bea3edf7b2c17f51e4 --- /dev/null +++ b/model-00005-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03e7bf18a4a68ec953316528fdb52b0713fb0c53cc89c71ed6b85b08e0307a00 +size 4828487128 diff --git a/model-00006-of-00044.safetensors b/model-00006-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fd56d73214f3703ab169832a01a79cb9d4f194e0 --- /dev/null +++ b/model-00006-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b00e374551d570ff85acf129fdd1ad38590e6ecbb00f4192635bfdf409b04d9 +size 4828487136 diff --git a/model-00007-of-00044.safetensors b/model-00007-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..2fdd0477e4ae468fdb8db7c95a71b85a6a47de88 --- /dev/null +++ b/model-00007-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fac61068e6b91837c76fbe6c914f1497a13d90703cdf787ccaaa381727a59dfc +size 4828487136 diff --git a/model-00008-of-00044.safetensors b/model-00008-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5cd374dc428681c509d6d15ae781e121a735e38b --- /dev/null +++ b/model-00008-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e8b7974dcfac90190760bc754ba486687236e96bd24a3fa01faf81b8aed07c4 +size 4828487136 diff --git a/model-00009-of-00044.safetensors b/model-00009-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f9f2e62ee92d4d84cd1a7bbae18bc32eeed5d06d --- /dev/null +++ b/model-00009-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:095c35d2945e5615e9e453b1b94deca83715ed96c4be0fb5d3a8bb9593a34fc1 +size 4828487136 diff --git a/model-00010-of-00044.safetensors b/model-00010-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..55adfdf1e3f1409ac0f2ab7458e43987f777ca37 --- /dev/null +++ b/model-00010-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f460d4b620d0d05ce57a433ee96f8167e13cf2f5e46c4c6569f4ee4f83f44b5 +size 4828487136 diff --git a/model-00011-of-00044.safetensors b/model-00011-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..29777157eb5c3f2ecfed950824e257fd24ed7ab0 --- /dev/null +++ b/model-00011-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e0c4748207dbe7a730b864ed55c6335ecfc52bb835efd0afc369a7f319fe8b5 +size 4828487136 diff --git a/model-00012-of-00044.safetensors b/model-00012-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c1377901becdac321da4a03f43fd67e3a2fe183e --- /dev/null +++ b/model-00012-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a983fb4f4e4033e011c6c81b442883e1426c49551a43531d3ca098d2eac3ff3 +size 4828487136 diff --git a/model-00013-of-00044.safetensors b/model-00013-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..20ff2935667d9fcb9e4c7603e4bf69a6ec98b1c1 --- /dev/null +++ b/model-00013-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f859c418f726af6a757f949e8fb6f4205bbd02560e6ca4fce89afc6f221f91a6 +size 4828487136 diff --git a/model-00014-of-00044.safetensors b/model-00014-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f864d62fcd7cad2ed097471d9de95645e0ad50a6 --- /dev/null +++ b/model-00014-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ec78fd18414df686b86871ff404b305e5418e7de9cc40ca9a3b291df3d8a21d +size 4828487136 diff --git a/model-00015-of-00044.safetensors b/model-00015-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..95ef65ee552f5cb2512cbcc23769d621bf053b41 --- /dev/null +++ b/model-00015-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40dd45dceed77b8c0020f663f1492b2f5894c0ebb4886d614c5983aa5d77591a +size 4828487136 diff --git a/model-00016-of-00044.safetensors b/model-00016-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d30cc403e1fe0b357351162cf5a48a0013dd3c67 --- /dev/null +++ b/model-00016-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8061b14be84e109333f5b5cbff93cb4934c6b98bea88d9b850e92f2cbfc31a7a +size 4828487136 diff --git a/model-00017-of-00044.safetensors b/model-00017-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..76d3f97966d0befb85acce1abbb3fd2e036389ad --- /dev/null +++ b/model-00017-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b752515de1890a278626e11e078ce3533d894713d26c084d579ed7ce2504fe05 +size 4828487136 diff --git a/model-00018-of-00044.safetensors b/model-00018-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..92fe35bdd8f6dcb798ffe4ea4b3225fcb844208d --- /dev/null +++ b/model-00018-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ea30a42d78d79fd4770ae1a4e6ebc51e170cfcdc4284c6dd0e642748c45cdbe +size 4828487136 diff --git a/model-00019-of-00044.safetensors b/model-00019-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..974f5dcc61b690af818e72adbf141d672cbfb68f --- /dev/null +++ b/model-00019-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09a8cb39f4176a2142bf70976a484fa3568b05e4c4fbff83ce95774ae5921043 +size 4828487136 diff --git a/model-00020-of-00044.safetensors b/model-00020-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..fb1abbc3052068153232d0df65a034d82aea5476 --- /dev/null +++ b/model-00020-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a2dc1ec004dfd62693511442564f2ad9e3ff1dbc48aaa4935ff9eb38e92b7f5 +size 4828487136 diff --git a/model-00021-of-00044.safetensors b/model-00021-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..22f178e6621038937169ac929f71996a42d4084d --- /dev/null +++ b/model-00021-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1545d0e0cced85060e4938fca8757a80edc693fb86b977d3d736dea28b6b3a7c +size 4828487136 diff --git a/model-00022-of-00044.safetensors b/model-00022-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f48abe26f1649e360573dc5eea020a041b8d0586 --- /dev/null +++ b/model-00022-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1d3b87e26db4fe86522a65c993e97b0d5eb036f54fa0f44f70a00bab84470525 +size 4828487136 diff --git a/model-00023-of-00044.safetensors b/model-00023-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..08692c6948a0baa7c21df19295a367160037c71a --- /dev/null +++ b/model-00023-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:427cc7e7e9c2a23019445b7dce9d98731d0415e6ee47aea1a61c687f1b5ec9a7 +size 4828487136 diff --git a/model-00024-of-00044.safetensors b/model-00024-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5069e1008fe681cc4e4ca2f5bd6b4e36fa3b1cc8 --- /dev/null +++ b/model-00024-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:070eb3f3256a6361c2909786c65e7ab8fde3a91db9198233a8bf5d1058cde8dc +size 4828487136 diff --git a/model-00025-of-00044.safetensors b/model-00025-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..177f72129a56887c79367ff8632ab34a94d3e640 --- /dev/null +++ b/model-00025-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f6525e0d90947a5425d96759b27404e1b4fce59415932922920bcc5daead70e +size 4828487136 diff --git a/model-00026-of-00044.safetensors b/model-00026-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..48215a80429e73aaa43cff66146c809ed00e3693 --- /dev/null +++ b/model-00026-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:59dac12b3e51f0cb0843337b2b1d5465f62dacdb94b433819b81275965982dec +size 4828487136 diff --git a/model-00027-of-00044.safetensors b/model-00027-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..5107a9e5023e9d5441db8b104765ab40b13aa034 --- /dev/null +++ b/model-00027-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee363eb2cbdda57fff7e0779e7d0a777a816162b8ef0b6fa4b4f7b8d7d020a72 +size 4828487136 diff --git a/model-00028-of-00044.safetensors b/model-00028-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..08ac2dc8e9455842f0846c396cf84f8510973d59 --- /dev/null +++ b/model-00028-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3be91c30d7103331e9db26ad98a8c8124acd0dc5714f4283e17a6b7fe1c86693 +size 4828487136 diff --git a/model-00029-of-00044.safetensors b/model-00029-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f7b097cc02da1de6c94e67c839a7455679d9a2d9 --- /dev/null +++ b/model-00029-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd14cb918c43b2929354c99eac53cb5454b9242db059f1b1b3588c7a86c93a73 +size 4828487136 diff --git a/model-00030-of-00044.safetensors b/model-00030-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..adb67adcbc6a59733d1c2798a58171fea5483240 --- /dev/null +++ b/model-00030-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:787b3b36d29dc542497b649588bc1ec6e6c095d8db258e2cfedbd65bb260c321 +size 4828487136 diff --git a/model-00031-of-00044.safetensors b/model-00031-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..73e0323e020a735246ee0d1ba9055c377c85e203 --- /dev/null +++ b/model-00031-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a5c9cf4c31851fd9b257704441f2e05138c72461550819e83e753e0997376a7 +size 4828487136 diff --git a/model-00032-of-00044.safetensors b/model-00032-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9c37117b06611f1b83d83e593e534c0ce66fb255 --- /dev/null +++ b/model-00032-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:174a7b0c25c54381e69c19227ef5bf8ea2c0bc78363df26e2cb6b4b655ab3a10 +size 4828487136 diff --git a/model-00033-of-00044.safetensors b/model-00033-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..f4a0c17488f26669d0ef80439222ea0f6665ecce --- /dev/null +++ b/model-00033-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e88ff0d10f57c7fb9355dc6737cfc8e3a1f31b5df615a72ab476c05def48afc3 +size 4828487136 diff --git a/model-00034-of-00044.safetensors b/model-00034-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..b31ee2a52f9e09713c0e07e5015bf1b776a5c615 --- /dev/null +++ b/model-00034-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84afaf085ed0d184036ca896cda0b8a95f29cb431ec7fedebafa30e4ea320c80 +size 4828487136 diff --git a/model-00035-of-00044.safetensors b/model-00035-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..a76c0bd92a690ffb7b56c3cacc04a6a64515d6d4 --- /dev/null +++ b/model-00035-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13544a252387c38cc6c34c4c4e8e8af7b1fd3692647972cb6523069e921a598b +size 4828487256 diff --git a/model-00036-of-00044.safetensors b/model-00036-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..63f4272e4338651d36d6b6bd581d0760719cc817 --- /dev/null +++ b/model-00036-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0db1ea30a1112fd84f28e3ca474efd0fc32af2febfa39fde18517532244ae3df +size 4828487272 diff --git a/model-00037-of-00044.safetensors b/model-00037-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..c30ad2c5639f8ecf822cf0d33ca3ac61296d0646 --- /dev/null +++ b/model-00037-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9e2bb141a601553026689ef0536cf50848ecfe75d1a5db4e05a1ae17d639fd05 +size 4828487272 diff --git a/model-00038-of-00044.safetensors b/model-00038-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..60ca4d96849f28c3d620f75ffb475a779564385d --- /dev/null +++ b/model-00038-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8b9aaaec761b0034a9567cc909e496acc4ea2dc7d5d6d62e4a6ef2b7926731d +size 4828487272 diff --git a/model-00039-of-00044.safetensors b/model-00039-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..d7ed6e74d09ec4377173b5591cd8bbdff575248e --- /dev/null +++ b/model-00039-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3364ac3eaf1c1dc5a57b425a6cda09c430b764d6f265770a6eef3c2a0611b98e +size 4828487272 diff --git a/model-00040-of-00044.safetensors b/model-00040-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..9872329fc8399d0f1c00e9e144d4370da64f2577 --- /dev/null +++ b/model-00040-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab1cbf788859a7410c88dfe1ee3d77315b8934602602dd2bf4c779ef86e0db37 +size 4828487272 diff --git a/model-00041-of-00044.safetensors b/model-00041-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..27dae0bf2247d37a9df0b7cdd4d7af55b7e02e4c --- /dev/null +++ b/model-00041-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08ab0a61d0a223a546ef6d11aa2a2f58556a7e325b34a8d9d9aa8039a8384eb9 +size 4828487272 diff --git a/model-00042-of-00044.safetensors b/model-00042-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..96a5aac7a4c5ca833d82560baf150109ba056dc9 --- /dev/null +++ b/model-00042-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1393292f9e6724c0210fade389941d5687bde1491c938c1e4ad22d9a8b582640 +size 4828487272 diff --git a/model-00043-of-00044.safetensors b/model-00043-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..0cf65ea363505a470d92d077890ef2bb762e63da --- /dev/null +++ b/model-00043-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f36c183e21e7b057e0473ad995943be82fd52e021eec2085308bd1d5527d1710 +size 4096020264 diff --git a/model-00044-of-00044.safetensors b/model-00044-of-00044.safetensors new file mode 100644 index 0000000000000000000000000000000000000000..189b2ba970360412785980a181e8aa7ae7624a18 --- /dev/null +++ b/model-00044-of-00044.safetensors @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71b079a02069ebe5a5f4f8135815f136afb1d6fc4423620dcb2a14fa399cdf12 +size 4202692736 diff --git a/model.safetensors.index.json b/model.safetensors.index.json new file mode 100644 index 0000000000000000000000000000000000000000..122fb3d2d0d5895f607c1d619dfa3a3536abecbb --- /dev/null +++ b/model.safetensors.index.json @@ -0,0 +1,5554 @@ +{ + "metadata": { + "total_size": 211201212416 + }, + "weight_map": { + "lm_head.weight": "model-00044-of-00044.safetensors", + "model.embed_tokens.weight": "model-00001-of-00044.safetensors", + "model.layers.0.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.down_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.indices": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.res_centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.gate_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.up_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.up_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.0.mlp.up_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.mlp.up_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.0.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.indices": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.res_centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.k_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.indices": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.res_centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.o_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.indices": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.res_centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.q_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.indices": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.perm": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.res_centroids.weight": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.weight_bias": "model-00001-of-00044.safetensors", + "model.layers.0.self_attn.v_proj.weight_scale": "model-00001-of-00044.safetensors", + "model.layers.1.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.down_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.gate_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.mlp.up_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.k_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.o_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.q_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.1.self_attn.v_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.10.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.down_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.gate_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.mlp.up_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.k_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.o_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.q_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.10.self_attn.v_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.100.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.down_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.gate_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.mlp.up_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.k_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.o_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.q_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.100.self_attn.v_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.down_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.gate_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.mlp.up_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.k_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.o_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.q_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.101.self_attn.v_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.down_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.gate_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.up_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.up_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.102.mlp.up_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.mlp.up_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.k_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.o_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.q_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.102.self_attn.v_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.103.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.down_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.gate_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.mlp.up_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.k_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.o_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.q_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.103.self_attn.v_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.input_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.down_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.gate_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.mlp.up_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.post_attention_layernorm.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.k_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.o_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.q_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.104.self_attn.v_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.down_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.gate_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.up_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.up_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.105.mlp.up_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.mlp.up_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.k_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.o_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.q_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.indices": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.perm": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.res_centroids.weight": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.weight_bias": "model-00036-of-00044.safetensors", + "model.layers.105.self_attn.v_proj.weight_scale": "model-00036-of-00044.safetensors", + "model.layers.106.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.down_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.gate_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.mlp.up_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.k_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.o_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.q_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.106.self_attn.v_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.input_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.down_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.gate_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.mlp.up_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.post_attention_layernorm.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.k_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.o_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.q_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.107.self_attn.v_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.down_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.gate_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.up_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.up_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.108.mlp.up_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.mlp.up_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.k_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.o_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.q_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.indices": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.perm": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.res_centroids.weight": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.weight_bias": "model-00037-of-00044.safetensors", + "model.layers.108.self_attn.v_proj.weight_scale": "model-00037-of-00044.safetensors", + "model.layers.109.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.down_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.gate_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.mlp.up_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.k_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.o_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.q_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.109.self_attn.v_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.11.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.down_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.gate_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.mlp.up_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.k_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.o_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.q_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.11.self_attn.v_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.110.input_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.down_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.gate_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.mlp.up_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.post_attention_layernorm.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.k_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.o_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.q_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.110.self_attn.v_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.down_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.gate_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.up_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.up_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.111.mlp.up_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.mlp.up_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.k_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.o_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.q_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.indices": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.perm": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.res_centroids.weight": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.weight_bias": "model-00038-of-00044.safetensors", + "model.layers.111.self_attn.v_proj.weight_scale": "model-00038-of-00044.safetensors", + "model.layers.112.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.down_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.gate_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.mlp.up_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.k_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.o_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.q_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.112.self_attn.v_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.input_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.down_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.gate_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.mlp.up_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.post_attention_layernorm.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.k_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.o_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.q_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.113.self_attn.v_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.down_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.gate_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.up_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.up_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.114.mlp.up_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.mlp.up_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.k_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.o_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.q_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.indices": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.perm": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.res_centroids.weight": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.weight_bias": "model-00039-of-00044.safetensors", + "model.layers.114.self_attn.v_proj.weight_scale": "model-00039-of-00044.safetensors", + "model.layers.115.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.down_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.gate_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.mlp.up_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.k_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.o_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.q_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.115.self_attn.v_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.input_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.down_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.gate_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.mlp.up_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.post_attention_layernorm.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.k_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.o_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.q_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.116.self_attn.v_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.down_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.gate_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.up_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.up_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.117.mlp.up_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.mlp.up_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.k_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.o_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.q_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.indices": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.perm": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.res_centroids.weight": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.weight_bias": "model-00040-of-00044.safetensors", + "model.layers.117.self_attn.v_proj.weight_scale": "model-00040-of-00044.safetensors", + "model.layers.118.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.down_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.gate_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.mlp.up_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.k_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.o_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.q_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.118.self_attn.v_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.input_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.down_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.gate_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.mlp.up_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.post_attention_layernorm.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.k_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.o_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.q_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.119.self_attn.v_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.12.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.down_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.gate_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.up_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.up_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.12.mlp.up_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.mlp.up_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.12.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.k_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.o_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.q_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.12.self_attn.v_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.120.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.down_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.gate_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.up_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.up_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.120.mlp.up_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.mlp.up_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.120.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.k_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.o_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.q_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.indices": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.perm": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.res_centroids.weight": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.weight_bias": "model-00041-of-00044.safetensors", + "model.layers.120.self_attn.v_proj.weight_scale": "model-00041-of-00044.safetensors", + "model.layers.121.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.down_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.gate_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.mlp.up_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.k_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.o_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.q_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.121.self_attn.v_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.input_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.down_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.gate_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.mlp.up_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.post_attention_layernorm.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.k_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.o_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.q_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.122.self_attn.v_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.down_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.gate_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.up_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.up_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.123.mlp.up_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.mlp.up_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.k_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.o_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.q_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.indices": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.perm": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.res_centroids.weight": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.weight_bias": "model-00042-of-00044.safetensors", + "model.layers.123.self_attn.v_proj.weight_scale": "model-00042-of-00044.safetensors", + "model.layers.124.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.down_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.gate_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.mlp.up_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.k_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.o_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.q_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.124.self_attn.v_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.input_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.down_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.gate_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.mlp.up_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.post_attention_layernorm.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.k_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.o_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.q_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.indices": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.perm": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.res_centroids.weight": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.weight_bias": "model-00043-of-00044.safetensors", + "model.layers.125.self_attn.v_proj.weight_scale": "model-00043-of-00044.safetensors", + "model.layers.13.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.down_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.gate_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.mlp.up_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.k_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.o_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.q_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.13.self_attn.v_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.input_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.down_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.gate_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.mlp.up_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.post_attention_layernorm.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.k_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.o_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.q_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.14.self_attn.v_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.down_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.gate_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.up_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.up_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.15.mlp.up_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.mlp.up_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.k_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.o_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.q_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.indices": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.perm": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.res_centroids.weight": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.weight_bias": "model-00006-of-00044.safetensors", + "model.layers.15.self_attn.v_proj.weight_scale": "model-00006-of-00044.safetensors", + "model.layers.16.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.down_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.gate_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.mlp.up_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.k_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.o_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.q_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.16.self_attn.v_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.input_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.down_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.gate_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.mlp.up_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.post_attention_layernorm.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.k_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.o_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.q_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.17.self_attn.v_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.down_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.gate_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.up_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.up_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.18.mlp.up_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.mlp.up_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.k_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.o_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.q_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.indices": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.perm": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.res_centroids.weight": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.weight_bias": "model-00007-of-00044.safetensors", + "model.layers.18.self_attn.v_proj.weight_scale": "model-00007-of-00044.safetensors", + "model.layers.19.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.down_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.gate_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.mlp.up_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.k_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.o_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.q_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.19.self_attn.v_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.2.input_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.down_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.gate_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.mlp.up_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.post_attention_layernorm.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.k_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.o_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.q_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.2.self_attn.v_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.20.input_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.down_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.gate_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.mlp.up_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.post_attention_layernorm.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.k_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.o_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.q_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.20.self_attn.v_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.down_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.gate_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.up_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.up_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.21.mlp.up_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.mlp.up_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.k_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.o_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.q_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.indices": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.perm": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.res_centroids.weight": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.weight_bias": "model-00008-of-00044.safetensors", + "model.layers.21.self_attn.v_proj.weight_scale": "model-00008-of-00044.safetensors", + "model.layers.22.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.down_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.gate_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.mlp.up_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.k_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.o_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.q_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.22.self_attn.v_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.input_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.down_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.gate_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.mlp.up_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.post_attention_layernorm.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.k_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.o_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.q_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.23.self_attn.v_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.down_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.gate_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.up_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.up_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.24.mlp.up_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.mlp.up_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.k_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.o_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.q_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.indices": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.perm": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.res_centroids.weight": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.weight_bias": "model-00009-of-00044.safetensors", + "model.layers.24.self_attn.v_proj.weight_scale": "model-00009-of-00044.safetensors", + "model.layers.25.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.down_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.gate_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.mlp.up_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.k_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.o_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.q_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.25.self_attn.v_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.input_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.down_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.gate_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.mlp.up_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.post_attention_layernorm.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.k_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.o_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.q_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.26.self_attn.v_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.down_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.gate_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.up_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.up_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.27.mlp.up_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.mlp.up_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.k_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.o_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.q_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.indices": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.perm": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.res_centroids.weight": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.weight_bias": "model-00010-of-00044.safetensors", + "model.layers.27.self_attn.v_proj.weight_scale": "model-00010-of-00044.safetensors", + "model.layers.28.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.down_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.gate_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.mlp.up_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.k_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.o_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.q_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.28.self_attn.v_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.input_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.down_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.gate_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.mlp.up_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.post_attention_layernorm.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.k_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.o_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.q_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.29.self_attn.v_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.3.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.down_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.gate_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.up_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.up_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.3.mlp.up_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.mlp.up_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.3.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.k_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.o_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.q_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.indices": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.perm": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.res_centroids.weight": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.weight_bias": "model-00002-of-00044.safetensors", + "model.layers.3.self_attn.v_proj.weight_scale": "model-00002-of-00044.safetensors", + "model.layers.30.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.down_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.gate_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.up_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.up_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.30.mlp.up_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.mlp.up_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.30.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.k_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.o_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.q_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.indices": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.perm": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.res_centroids.weight": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.weight_bias": "model-00011-of-00044.safetensors", + "model.layers.30.self_attn.v_proj.weight_scale": "model-00011-of-00044.safetensors", + "model.layers.31.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.down_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.gate_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.mlp.up_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.k_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.o_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.q_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.31.self_attn.v_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.input_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.down_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.gate_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.mlp.up_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.post_attention_layernorm.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.k_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.o_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.q_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.32.self_attn.v_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.down_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.gate_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.up_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.up_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.33.mlp.up_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.mlp.up_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.k_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.o_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.q_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.indices": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.perm": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.res_centroids.weight": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.weight_bias": "model-00012-of-00044.safetensors", + "model.layers.33.self_attn.v_proj.weight_scale": "model-00012-of-00044.safetensors", + "model.layers.34.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.down_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.gate_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.mlp.up_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.k_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.o_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.q_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.34.self_attn.v_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.input_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.down_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.gate_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.mlp.up_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.post_attention_layernorm.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.k_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.o_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.q_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.35.self_attn.v_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.down_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.gate_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.up_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.up_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.36.mlp.up_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.mlp.up_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.k_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.o_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.q_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.indices": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.perm": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.res_centroids.weight": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.weight_bias": "model-00013-of-00044.safetensors", + "model.layers.36.self_attn.v_proj.weight_scale": "model-00013-of-00044.safetensors", + "model.layers.37.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.down_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.gate_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.mlp.up_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.k_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.o_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.q_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.37.self_attn.v_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.input_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.down_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.gate_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.mlp.up_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.post_attention_layernorm.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.k_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.o_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.q_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.38.self_attn.v_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.down_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.gate_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.up_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.up_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.39.mlp.up_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.mlp.up_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.k_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.o_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.q_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.indices": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.perm": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.res_centroids.weight": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.weight_bias": "model-00014-of-00044.safetensors", + "model.layers.39.self_attn.v_proj.weight_scale": "model-00014-of-00044.safetensors", + "model.layers.4.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.down_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.gate_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.mlp.up_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.k_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.o_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.q_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.4.self_attn.v_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.40.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.down_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.gate_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.mlp.up_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.k_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.o_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.q_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.40.self_attn.v_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.input_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.down_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.gate_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.mlp.up_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.post_attention_layernorm.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.k_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.o_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.q_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.41.self_attn.v_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.down_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.gate_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.up_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.up_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.42.mlp.up_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.mlp.up_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.k_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.o_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.q_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.indices": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.perm": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.res_centroids.weight": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.weight_bias": "model-00015-of-00044.safetensors", + "model.layers.42.self_attn.v_proj.weight_scale": "model-00015-of-00044.safetensors", + "model.layers.43.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.down_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.gate_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.mlp.up_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.k_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.o_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.q_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.43.self_attn.v_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.input_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.down_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.gate_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.mlp.up_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.post_attention_layernorm.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.k_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.o_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.q_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.44.self_attn.v_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.down_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.gate_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.up_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.up_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.45.mlp.up_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.mlp.up_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.k_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.o_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.q_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.indices": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.perm": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.res_centroids.weight": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.weight_bias": "model-00016-of-00044.safetensors", + "model.layers.45.self_attn.v_proj.weight_scale": "model-00016-of-00044.safetensors", + "model.layers.46.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.down_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.gate_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.mlp.up_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.k_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.o_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.q_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.46.self_attn.v_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.input_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.down_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.gate_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.mlp.up_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.post_attention_layernorm.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.k_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.o_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.q_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.47.self_attn.v_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.down_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.gate_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.up_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.up_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.48.mlp.up_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.mlp.up_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.k_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.o_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.q_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.indices": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.perm": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.res_centroids.weight": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.weight_bias": "model-00017-of-00044.safetensors", + "model.layers.48.self_attn.v_proj.weight_scale": "model-00017-of-00044.safetensors", + "model.layers.49.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.down_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.gate_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.mlp.up_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.k_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.o_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.q_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.49.self_attn.v_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.5.input_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.down_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.gate_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.mlp.up_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.post_attention_layernorm.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.k_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.o_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.q_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.5.self_attn.v_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.50.input_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.down_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.gate_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.mlp.up_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.post_attention_layernorm.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.k_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.o_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.q_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.50.self_attn.v_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.down_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.gate_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.up_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.up_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.51.mlp.up_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.mlp.up_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.k_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.o_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.q_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.indices": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.perm": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.res_centroids.weight": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.weight_bias": "model-00018-of-00044.safetensors", + "model.layers.51.self_attn.v_proj.weight_scale": "model-00018-of-00044.safetensors", + "model.layers.52.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.down_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.gate_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.mlp.up_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.k_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.o_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.q_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.52.self_attn.v_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.input_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.down_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.gate_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.mlp.up_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.post_attention_layernorm.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.k_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.o_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.q_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.53.self_attn.v_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.down_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.gate_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.up_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.up_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.54.mlp.up_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.mlp.up_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.k_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.o_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.q_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.indices": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.perm": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.res_centroids.weight": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.weight_bias": "model-00019-of-00044.safetensors", + "model.layers.54.self_attn.v_proj.weight_scale": "model-00019-of-00044.safetensors", + "model.layers.55.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.down_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.gate_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.mlp.up_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.k_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.o_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.q_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.55.self_attn.v_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.input_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.down_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.gate_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.mlp.up_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.post_attention_layernorm.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.k_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.o_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.q_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.56.self_attn.v_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.down_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.gate_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.up_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.up_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.57.mlp.up_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.mlp.up_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.k_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.o_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.q_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.indices": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.perm": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.res_centroids.weight": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.weight_bias": "model-00020-of-00044.safetensors", + "model.layers.57.self_attn.v_proj.weight_scale": "model-00020-of-00044.safetensors", + "model.layers.58.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.down_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.gate_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.mlp.up_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.k_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.o_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.q_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.58.self_attn.v_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.input_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.down_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.gate_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.mlp.up_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.post_attention_layernorm.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.k_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.o_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.q_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.59.self_attn.v_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.6.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.down_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.gate_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.up_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.up_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.6.mlp.up_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.mlp.up_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.6.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.k_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.o_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.q_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.indices": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.perm": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.res_centroids.weight": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.weight_bias": "model-00003-of-00044.safetensors", + "model.layers.6.self_attn.v_proj.weight_scale": "model-00003-of-00044.safetensors", + "model.layers.60.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.down_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.gate_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.up_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.up_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.60.mlp.up_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.mlp.up_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.60.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.k_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.o_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.q_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.indices": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.perm": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.res_centroids.weight": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.weight_bias": "model-00021-of-00044.safetensors", + "model.layers.60.self_attn.v_proj.weight_scale": "model-00021-of-00044.safetensors", + "model.layers.61.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.down_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.gate_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.mlp.up_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.k_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.o_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.q_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.61.self_attn.v_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.input_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.down_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.gate_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.mlp.up_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.post_attention_layernorm.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.k_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.o_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.q_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.62.self_attn.v_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.down_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.gate_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.up_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.up_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.63.mlp.up_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.mlp.up_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.k_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.o_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.q_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.indices": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.perm": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.res_centroids.weight": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.weight_bias": "model-00022-of-00044.safetensors", + "model.layers.63.self_attn.v_proj.weight_scale": "model-00022-of-00044.safetensors", + "model.layers.64.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.down_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.gate_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.mlp.up_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.k_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.o_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.q_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.64.self_attn.v_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.input_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.down_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.gate_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.mlp.up_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.post_attention_layernorm.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.k_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.o_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.q_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.65.self_attn.v_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.down_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.gate_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.up_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.up_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.66.mlp.up_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.mlp.up_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.k_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.o_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.q_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.indices": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.perm": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.res_centroids.weight": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.weight_bias": "model-00023-of-00044.safetensors", + "model.layers.66.self_attn.v_proj.weight_scale": "model-00023-of-00044.safetensors", + "model.layers.67.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.down_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.gate_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.mlp.up_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.k_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.o_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.q_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.67.self_attn.v_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.input_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.down_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.gate_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.mlp.up_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.post_attention_layernorm.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.k_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.o_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.q_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.68.self_attn.v_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.down_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.gate_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.up_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.up_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.69.mlp.up_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.mlp.up_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.k_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.o_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.q_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.indices": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.perm": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.res_centroids.weight": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.weight_bias": "model-00024-of-00044.safetensors", + "model.layers.69.self_attn.v_proj.weight_scale": "model-00024-of-00044.safetensors", + "model.layers.7.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.down_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.gate_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.mlp.up_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.k_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.o_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.q_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.7.self_attn.v_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.70.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.down_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.gate_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.mlp.up_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.k_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.o_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.q_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.70.self_attn.v_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.input_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.down_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.gate_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.mlp.up_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.post_attention_layernorm.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.k_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.o_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.q_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.71.self_attn.v_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.down_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.gate_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.up_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.up_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.72.mlp.up_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.mlp.up_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.k_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.o_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.q_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.indices": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.perm": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.res_centroids.weight": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.weight_bias": "model-00025-of-00044.safetensors", + "model.layers.72.self_attn.v_proj.weight_scale": "model-00025-of-00044.safetensors", + "model.layers.73.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.down_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.gate_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.mlp.up_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.k_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.o_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.q_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.73.self_attn.v_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.input_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.down_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.gate_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.mlp.up_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.post_attention_layernorm.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.k_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.o_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.q_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.74.self_attn.v_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.down_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.gate_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.up_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.up_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.75.mlp.up_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.mlp.up_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.k_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.o_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.q_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.indices": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.perm": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.res_centroids.weight": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.weight_bias": "model-00026-of-00044.safetensors", + "model.layers.75.self_attn.v_proj.weight_scale": "model-00026-of-00044.safetensors", + "model.layers.76.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.down_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.gate_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.mlp.up_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.k_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.o_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.q_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.76.self_attn.v_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.input_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.down_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.gate_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.mlp.up_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.post_attention_layernorm.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.k_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.o_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.q_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.77.self_attn.v_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.down_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.gate_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.up_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.up_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.78.mlp.up_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.mlp.up_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.k_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.o_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.q_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.indices": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.perm": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.res_centroids.weight": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.weight_bias": "model-00027-of-00044.safetensors", + "model.layers.78.self_attn.v_proj.weight_scale": "model-00027-of-00044.safetensors", + "model.layers.79.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.down_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.gate_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.mlp.up_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.k_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.o_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.q_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.79.self_attn.v_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.8.input_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.down_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.gate_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.mlp.up_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.post_attention_layernorm.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.k_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.o_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.q_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.8.self_attn.v_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.80.input_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.down_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.gate_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.mlp.up_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.post_attention_layernorm.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.k_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.o_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.q_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.80.self_attn.v_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.down_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.gate_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.up_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.up_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.81.mlp.up_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.mlp.up_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.k_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.o_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.q_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.indices": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.perm": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.res_centroids.weight": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.weight_bias": "model-00028-of-00044.safetensors", + "model.layers.81.self_attn.v_proj.weight_scale": "model-00028-of-00044.safetensors", + "model.layers.82.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.down_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.gate_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.mlp.up_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.k_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.o_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.q_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.82.self_attn.v_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.input_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.down_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.gate_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.mlp.up_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.post_attention_layernorm.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.k_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.o_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.q_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.83.self_attn.v_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.down_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.gate_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.up_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.up_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.84.mlp.up_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.mlp.up_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.k_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.o_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.q_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.indices": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.perm": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.res_centroids.weight": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.weight_bias": "model-00029-of-00044.safetensors", + "model.layers.84.self_attn.v_proj.weight_scale": "model-00029-of-00044.safetensors", + "model.layers.85.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.down_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.gate_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.mlp.up_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.k_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.o_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.q_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.85.self_attn.v_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.input_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.down_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.gate_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.mlp.up_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.post_attention_layernorm.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.k_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.o_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.q_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.86.self_attn.v_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.down_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.gate_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.up_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.up_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.87.mlp.up_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.mlp.up_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.k_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.o_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.q_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.indices": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.perm": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.res_centroids.weight": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.weight_bias": "model-00030-of-00044.safetensors", + "model.layers.87.self_attn.v_proj.weight_scale": "model-00030-of-00044.safetensors", + "model.layers.88.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.down_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.gate_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.mlp.up_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.k_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.o_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.q_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.88.self_attn.v_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.input_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.down_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.gate_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.mlp.up_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.post_attention_layernorm.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.k_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.o_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.q_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.89.self_attn.v_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.9.input_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.perm": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.weight_bias": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.down_proj.weight_scale": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.gate_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.up_proj.centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.indices": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.up_proj.res_centroids.weight": "model-00005-of-00044.safetensors", + "model.layers.9.mlp.up_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.mlp.up_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.9.post_attention_layernorm.weight": "model-00005-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.k_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.o_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.q_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.indices": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.perm": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.res_centroids.weight": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.weight_bias": "model-00004-of-00044.safetensors", + "model.layers.9.self_attn.v_proj.weight_scale": "model-00004-of-00044.safetensors", + "model.layers.90.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.down_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.gate_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.up_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.up_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.90.mlp.up_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.mlp.up_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.90.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.k_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.o_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.q_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.indices": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.perm": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.res_centroids.weight": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.weight_bias": "model-00031-of-00044.safetensors", + "model.layers.90.self_attn.v_proj.weight_scale": "model-00031-of-00044.safetensors", + "model.layers.91.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.down_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.gate_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.mlp.up_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.k_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.o_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.q_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.91.self_attn.v_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.input_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.down_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.gate_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.mlp.up_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.post_attention_layernorm.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.k_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.o_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.q_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.92.self_attn.v_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.down_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.gate_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.up_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.up_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.93.mlp.up_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.mlp.up_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.k_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.o_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.q_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.indices": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.perm": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.res_centroids.weight": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.weight_bias": "model-00032-of-00044.safetensors", + "model.layers.93.self_attn.v_proj.weight_scale": "model-00032-of-00044.safetensors", + "model.layers.94.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.down_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.gate_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.mlp.up_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.k_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.o_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.q_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.94.self_attn.v_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.input_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.down_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.gate_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.mlp.up_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.post_attention_layernorm.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.k_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.o_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.q_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.95.self_attn.v_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.down_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.gate_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.up_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.up_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.96.mlp.up_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.mlp.up_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.k_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.o_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.q_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.indices": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.perm": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.res_centroids.weight": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.weight_bias": "model-00033-of-00044.safetensors", + "model.layers.96.self_attn.v_proj.weight_scale": "model-00033-of-00044.safetensors", + "model.layers.97.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.down_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.gate_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.mlp.up_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.k_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.o_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.q_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.97.self_attn.v_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.input_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.down_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.gate_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.mlp.up_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.post_attention_layernorm.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.k_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.o_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.q_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.98.self_attn.v_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.input_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.perm": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.weight_bias": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.down_proj.weight_scale": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.gate_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.up_proj.centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.indices": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.up_proj.res_centroids.weight": "model-00035-of-00044.safetensors", + "model.layers.99.mlp.up_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.mlp.up_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.post_attention_layernorm.weight": "model-00035-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.k_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.o_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.q_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.indices": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.perm": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.res_centroids.weight": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.weight_bias": "model-00034-of-00044.safetensors", + "model.layers.99.self_attn.v_proj.weight_scale": "model-00034-of-00044.safetensors", + "model.norm.weight": "model-00043-of-00044.safetensors" + } +} diff --git a/special_tokens_map.json b/special_tokens_map.json new file mode 100644 index 0000000000000000000000000000000000000000..02ee80b6196926a5ad790a004d9efd6ab1ba6542 --- /dev/null +++ b/special_tokens_map.json @@ -0,0 +1,16 @@ +{ + "bos_token": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false + }, + "eos_token": { + "content": "<|eot_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false + } +} diff --git a/tokenizer.json b/tokenizer.json new file mode 100644 index 0000000000000000000000000000000000000000..1c1d8d5c9024994f1d3b00f9662b8dd89ca13cf2 --- /dev/null +++ b/tokenizer.json @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b9e4e7fb171f92fd137b777cc2714bf87d11576700a1dcd7a399e7bbe39537b +size 17209920 diff --git a/tokenizer_config.json b/tokenizer_config.json new file mode 100644 index 0000000000000000000000000000000000000000..dc8720a277cf8d9aa312054ae42aab507cdf5dfb --- /dev/null +++ b/tokenizer_config.json @@ -0,0 +1,2063 @@ +{ + "added_tokens_decoder": { + "128000": { + "content": "<|begin_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128001": { + "content": "<|end_of_text|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128002": { + "content": "<|reserved_special_token_0|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128003": { + "content": "<|reserved_special_token_1|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128004": { + "content": "<|finetune_right_pad_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128005": { + "content": "<|reserved_special_token_2|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128006": { + "content": "<|start_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128007": { + "content": "<|end_header_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128008": { + "content": "<|eom_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128009": { + "content": "<|eot_id|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128010": { + "content": "<|python_tag|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128011": { + "content": "<|reserved_special_token_3|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128012": { + "content": "<|reserved_special_token_4|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128013": { + "content": "<|reserved_special_token_5|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128014": { + "content": "<|reserved_special_token_6|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128015": { + "content": "<|reserved_special_token_7|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128016": { + "content": "<|reserved_special_token_8|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128017": { + "content": "<|reserved_special_token_9|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128018": { + "content": "<|reserved_special_token_10|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128019": { + "content": "<|reserved_special_token_11|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128020": { + "content": "<|reserved_special_token_12|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128021": { + "content": "<|reserved_special_token_13|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128022": { + "content": "<|reserved_special_token_14|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128023": { + "content": "<|reserved_special_token_15|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128024": { + "content": "<|reserved_special_token_16|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128025": { + "content": "<|reserved_special_token_17|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128026": { + "content": "<|reserved_special_token_18|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128027": { + "content": "<|reserved_special_token_19|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128028": { + "content": "<|reserved_special_token_20|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128029": { + "content": "<|reserved_special_token_21|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128030": { + "content": "<|reserved_special_token_22|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128031": { + "content": "<|reserved_special_token_23|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128032": { + "content": "<|reserved_special_token_24|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128033": { + "content": "<|reserved_special_token_25|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128034": { + "content": "<|reserved_special_token_26|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128035": { + "content": "<|reserved_special_token_27|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128036": { + "content": "<|reserved_special_token_28|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128037": { + "content": "<|reserved_special_token_29|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128038": { + "content": "<|reserved_special_token_30|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128039": { + "content": "<|reserved_special_token_31|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128040": { + "content": "<|reserved_special_token_32|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128041": { + "content": "<|reserved_special_token_33|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128042": { + "content": "<|reserved_special_token_34|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128043": { + "content": "<|reserved_special_token_35|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128044": { + "content": "<|reserved_special_token_36|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128045": { + "content": "<|reserved_special_token_37|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128046": { + "content": "<|reserved_special_token_38|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128047": { + "content": "<|reserved_special_token_39|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128048": { + "content": "<|reserved_special_token_40|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128049": { + "content": "<|reserved_special_token_41|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128050": { + "content": "<|reserved_special_token_42|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128051": { + "content": "<|reserved_special_token_43|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128052": { + "content": "<|reserved_special_token_44|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128053": { + "content": "<|reserved_special_token_45|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128054": { + "content": "<|reserved_special_token_46|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128055": { + "content": "<|reserved_special_token_47|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128056": { + "content": "<|reserved_special_token_48|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128057": { + "content": "<|reserved_special_token_49|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128058": { + "content": "<|reserved_special_token_50|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128059": { + "content": "<|reserved_special_token_51|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128060": { + "content": "<|reserved_special_token_52|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128061": { + "content": "<|reserved_special_token_53|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128062": { + "content": "<|reserved_special_token_54|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128063": { + "content": "<|reserved_special_token_55|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128064": { + "content": "<|reserved_special_token_56|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128065": { + "content": "<|reserved_special_token_57|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128066": { + "content": "<|reserved_special_token_58|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128067": { + "content": "<|reserved_special_token_59|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128068": { + "content": "<|reserved_special_token_60|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128069": { + "content": "<|reserved_special_token_61|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128070": { + "content": "<|reserved_special_token_62|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128071": { + "content": "<|reserved_special_token_63|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128072": { + "content": "<|reserved_special_token_64|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128073": { + "content": "<|reserved_special_token_65|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128074": { + "content": "<|reserved_special_token_66|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128075": { + "content": "<|reserved_special_token_67|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128076": { + "content": "<|reserved_special_token_68|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128077": { + "content": "<|reserved_special_token_69|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128078": { + "content": "<|reserved_special_token_70|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128079": { + "content": "<|reserved_special_token_71|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128080": { + "content": "<|reserved_special_token_72|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128081": { + "content": "<|reserved_special_token_73|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128082": { + "content": "<|reserved_special_token_74|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128083": { + "content": "<|reserved_special_token_75|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128084": { + "content": "<|reserved_special_token_76|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128085": { + "content": "<|reserved_special_token_77|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128086": { + "content": "<|reserved_special_token_78|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128087": { + "content": "<|reserved_special_token_79|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128088": { + "content": "<|reserved_special_token_80|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128089": { + "content": "<|reserved_special_token_81|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128090": { + "content": "<|reserved_special_token_82|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128091": { + "content": "<|reserved_special_token_83|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128092": { + "content": "<|reserved_special_token_84|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128093": { + "content": "<|reserved_special_token_85|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128094": { + "content": "<|reserved_special_token_86|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128095": { + "content": "<|reserved_special_token_87|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128096": { + "content": "<|reserved_special_token_88|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128097": { + "content": "<|reserved_special_token_89|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128098": { + "content": "<|reserved_special_token_90|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128099": { + "content": "<|reserved_special_token_91|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128100": { + "content": "<|reserved_special_token_92|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128101": { + "content": "<|reserved_special_token_93|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128102": { + "content": "<|reserved_special_token_94|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128103": { + "content": "<|reserved_special_token_95|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128104": { + "content": "<|reserved_special_token_96|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128105": { + "content": "<|reserved_special_token_97|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128106": { + "content": "<|reserved_special_token_98|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128107": { + "content": "<|reserved_special_token_99|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128108": { + "content": "<|reserved_special_token_100|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128109": { + "content": "<|reserved_special_token_101|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128110": { + "content": "<|reserved_special_token_102|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128111": { + "content": "<|reserved_special_token_103|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128112": { + "content": "<|reserved_special_token_104|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128113": { + "content": "<|reserved_special_token_105|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128114": { + "content": "<|reserved_special_token_106|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128115": { + "content": "<|reserved_special_token_107|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128116": { + "content": "<|reserved_special_token_108|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128117": { + "content": "<|reserved_special_token_109|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128118": { + "content": "<|reserved_special_token_110|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128119": { + "content": "<|reserved_special_token_111|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128120": { + "content": "<|reserved_special_token_112|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128121": { + "content": "<|reserved_special_token_113|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128122": { + "content": "<|reserved_special_token_114|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128123": { + "content": "<|reserved_special_token_115|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128124": { + "content": "<|reserved_special_token_116|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128125": { + "content": "<|reserved_special_token_117|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128126": { + "content": "<|reserved_special_token_118|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128127": { + "content": "<|reserved_special_token_119|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128128": { + "content": "<|reserved_special_token_120|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128129": { + "content": "<|reserved_special_token_121|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128130": { + "content": "<|reserved_special_token_122|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128131": { + "content": "<|reserved_special_token_123|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128132": { + "content": "<|reserved_special_token_124|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128133": { + "content": "<|reserved_special_token_125|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128134": { + "content": "<|reserved_special_token_126|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128135": { + "content": "<|reserved_special_token_127|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128136": { + "content": "<|reserved_special_token_128|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128137": { + "content": "<|reserved_special_token_129|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128138": { + "content": "<|reserved_special_token_130|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128139": { + "content": "<|reserved_special_token_131|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128140": { + "content": "<|reserved_special_token_132|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128141": { + "content": "<|reserved_special_token_133|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128142": { + "content": "<|reserved_special_token_134|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128143": { + "content": "<|reserved_special_token_135|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128144": { + "content": "<|reserved_special_token_136|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128145": { + "content": "<|reserved_special_token_137|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128146": { + "content": "<|reserved_special_token_138|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128147": { + "content": "<|reserved_special_token_139|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128148": { + "content": "<|reserved_special_token_140|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128149": { + "content": "<|reserved_special_token_141|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128150": { + "content": "<|reserved_special_token_142|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128151": { + "content": "<|reserved_special_token_143|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128152": { + "content": "<|reserved_special_token_144|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128153": { + "content": "<|reserved_special_token_145|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128154": { + "content": "<|reserved_special_token_146|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128155": { + "content": "<|reserved_special_token_147|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128156": { + "content": "<|reserved_special_token_148|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128157": { + "content": "<|reserved_special_token_149|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128158": { + "content": "<|reserved_special_token_150|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128159": { + "content": "<|reserved_special_token_151|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128160": { + "content": "<|reserved_special_token_152|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128161": { + "content": "<|reserved_special_token_153|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128162": { + "content": "<|reserved_special_token_154|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128163": { + "content": "<|reserved_special_token_155|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128164": { + "content": "<|reserved_special_token_156|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128165": { + "content": "<|reserved_special_token_157|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128166": { + "content": "<|reserved_special_token_158|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128167": { + "content": "<|reserved_special_token_159|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128168": { + "content": "<|reserved_special_token_160|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128169": { + "content": "<|reserved_special_token_161|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128170": { + "content": "<|reserved_special_token_162|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128171": { + "content": "<|reserved_special_token_163|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128172": { + "content": "<|reserved_special_token_164|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128173": { + "content": "<|reserved_special_token_165|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128174": { + "content": "<|reserved_special_token_166|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128175": { + "content": "<|reserved_special_token_167|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128176": { + "content": "<|reserved_special_token_168|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128177": { + "content": "<|reserved_special_token_169|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128178": { + "content": "<|reserved_special_token_170|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128179": { + "content": "<|reserved_special_token_171|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128180": { + "content": "<|reserved_special_token_172|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128181": { + "content": "<|reserved_special_token_173|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128182": { + "content": "<|reserved_special_token_174|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128183": { + "content": "<|reserved_special_token_175|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128184": { + "content": "<|reserved_special_token_176|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128185": { + "content": "<|reserved_special_token_177|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128186": { + "content": "<|reserved_special_token_178|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128187": { + "content": "<|reserved_special_token_179|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128188": { + "content": "<|reserved_special_token_180|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128189": { + "content": "<|reserved_special_token_181|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128190": { + "content": "<|reserved_special_token_182|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128191": { + "content": "<|reserved_special_token_183|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128192": { + "content": "<|reserved_special_token_184|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128193": { + "content": "<|reserved_special_token_185|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128194": { + "content": "<|reserved_special_token_186|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128195": { + "content": "<|reserved_special_token_187|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128196": { + "content": "<|reserved_special_token_188|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128197": { + "content": "<|reserved_special_token_189|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128198": { + "content": "<|reserved_special_token_190|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128199": { + "content": "<|reserved_special_token_191|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128200": { + "content": "<|reserved_special_token_192|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128201": { + "content": "<|reserved_special_token_193|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128202": { + "content": "<|reserved_special_token_194|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128203": { + "content": "<|reserved_special_token_195|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128204": { + "content": "<|reserved_special_token_196|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128205": { + "content": "<|reserved_special_token_197|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128206": { + "content": "<|reserved_special_token_198|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128207": { + "content": "<|reserved_special_token_199|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128208": { + "content": "<|reserved_special_token_200|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128209": { + "content": "<|reserved_special_token_201|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128210": { + "content": "<|reserved_special_token_202|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128211": { + "content": "<|reserved_special_token_203|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128212": { + "content": "<|reserved_special_token_204|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128213": { + "content": "<|reserved_special_token_205|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128214": { + "content": "<|reserved_special_token_206|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128215": { + "content": "<|reserved_special_token_207|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128216": { + "content": "<|reserved_special_token_208|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128217": { + "content": "<|reserved_special_token_209|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128218": { + "content": "<|reserved_special_token_210|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128219": { + "content": "<|reserved_special_token_211|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128220": { + "content": "<|reserved_special_token_212|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128221": { + "content": "<|reserved_special_token_213|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128222": { + "content": "<|reserved_special_token_214|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128223": { + "content": "<|reserved_special_token_215|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128224": { + "content": "<|reserved_special_token_216|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128225": { + "content": "<|reserved_special_token_217|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128226": { + "content": "<|reserved_special_token_218|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128227": { + "content": "<|reserved_special_token_219|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128228": { + "content": "<|reserved_special_token_220|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128229": { + "content": "<|reserved_special_token_221|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128230": { + "content": "<|reserved_special_token_222|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128231": { + "content": "<|reserved_special_token_223|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128232": { + "content": "<|reserved_special_token_224|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128233": { + "content": "<|reserved_special_token_225|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128234": { + "content": "<|reserved_special_token_226|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128235": { + "content": "<|reserved_special_token_227|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128236": { + "content": "<|reserved_special_token_228|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128237": { + "content": "<|reserved_special_token_229|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128238": { + "content": "<|reserved_special_token_230|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128239": { + "content": "<|reserved_special_token_231|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128240": { + "content": "<|reserved_special_token_232|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128241": { + "content": "<|reserved_special_token_233|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128242": { + "content": "<|reserved_special_token_234|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128243": { + "content": "<|reserved_special_token_235|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128244": { + "content": "<|reserved_special_token_236|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128245": { + "content": "<|reserved_special_token_237|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128246": { + "content": "<|reserved_special_token_238|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128247": { + "content": "<|reserved_special_token_239|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128248": { + "content": "<|reserved_special_token_240|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128249": { + "content": "<|reserved_special_token_241|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128250": { + "content": "<|reserved_special_token_242|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128251": { + "content": "<|reserved_special_token_243|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128252": { + "content": "<|reserved_special_token_244|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128253": { + "content": "<|reserved_special_token_245|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128254": { + "content": "<|reserved_special_token_246|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + }, + "128255": { + "content": "<|reserved_special_token_247|>", + "lstrip": false, + "normalized": false, + "rstrip": false, + "single_word": false, + "special": true + } + }, + "bos_token": "<|begin_of_text|>", + "chat_template": "{{- bos_token }}\n{%- if custom_tools is defined %}\n {%- set tools = custom_tools %}\n{%- endif %}\n{%- if not tools_in_user_message is defined %}\n {%- set tools_in_user_message = true %}\n{%- endif %}\n{%- if not date_string is defined %}\n {%- set date_string = \"26 Jul 2024\" %}\n{%- endif %}\n{%- if not tools is defined %}\n {%- set tools = none %}\n{%- endif %}\n\n{#- This block extracts the system message, so we can slot it into the right place. #}\n{%- if messages[0]['role'] == 'system' %}\n {%- set system_message = messages[0]['content']|trim %}\n {%- set messages = messages[1:] %}\n{%- else %}\n {%- set system_message = \"\" %}\n{%- endif %}\n\n{#- System message + builtin tools #}\n{{- \"<|start_header_id|>system<|end_header_id|>\\n\\n\" }}\n{%- if builtin_tools is defined or tools is not none %}\n {{- \"Environment: ipython\\n\" }}\n{%- endif %}\n{%- if builtin_tools is defined %}\n {{- \"Tools: \" + builtin_tools | reject('equalto', 'code_interpreter') | join(\", \") + \"\\n\\n\"}}\n{%- endif %}\n{{- \"Cutting Knowledge Date: December 2023\\n\" }}\n{{- \"Today Date: \" + date_string + \"\\n\\n\" }}\n{%- if tools is not none and not tools_in_user_message %}\n {{- \"You have access to the following functions. To call a function, please respond with JSON for a function call.\" }}\n {{- 'Respond in the format {\"name\": function name, \"parameters\": dictionary of argument name and its value}.' }}\n {{- \"Do not use variables.\\n\\n\" }}\n {%- for t in tools %}\n {{- t | tojson(indent=4) }}\n {{- \"\\n\\n\" }}\n {%- endfor %}\n{%- endif %}\n{{- system_message }}\n{{- \"<|eot_id|>\" }}\n\n{#- Custom tools are passed in a user message with some extra guidance #}\n{%- if tools_in_user_message and not tools is none %}\n {#- Extract the first user message so we can plug it in here #}\n {%- if messages | length != 0 %}\n {%- set first_user_message = messages[0]['content']|trim %}\n {%- set messages = messages[1:] %}\n {%- else %}\n {{- raise_exception(\"Cannot put tools in the first user message when there's no first user message!\") }}\n{%- endif %}\n {{- '<|start_header_id|>user<|end_header_id|>\\n\\n' -}}\n {{- \"Given the following functions, please respond with a JSON for a function call \" }}\n {{- \"with its proper arguments that best answers the given prompt.\\n\\n\" }}\n {{- 'Respond in the format {\"name\": function name, \"parameters\": dictionary of argument name and its value}.' }}\n {{- \"Do not use variables.\\n\\n\" }}\n {%- for t in tools %}\n {{- t | tojson(indent=4) }}\n {{- \"\\n\\n\" }}\n {%- endfor %}\n {{- first_user_message + \"<|eot_id|>\"}}\n{%- endif %}\n\n{%- for message in messages %}\n {%- if not (message.role == 'ipython' or message.role == 'tool' or 'tool_calls' in message) %}\n {{- '<|start_header_id|>' + message['role'] + '<|end_header_id|>\\n\\n'+ message['content'] | trim + '<|eot_id|>' }}\n {%- elif 'tool_calls' in message %}\n {%- if not message.tool_calls|length == 1 %}\n {{- raise_exception(\"This model only supports single tool-calls at once!\") }}\n {%- endif %}\n {%- set tool_call = message.tool_calls[0].function %}\n {%- if builtin_tools is defined and tool_call.name in builtin_tools %}\n {{- '<|start_header_id|>assistant<|end_header_id|>\\n\\n' -}}\n {{- \"<|python_tag|>\" + tool_call.name + \".call(\" }}\n {%- for arg_name, arg_val in tool_call.arguments | items %}\n {{- arg_name + '=\"' + arg_val + '\"' }}\n {%- if not loop.last %}\n {{- \", \" }}\n {%- endif %}\n {%- endfor %}\n {{- \")\" }}\n {%- else %}\n {{- '<|start_header_id|>assistant<|end_header_id|>\\n\\n' -}}\n {{- '{\"name\": \"' + tool_call.name + '\", ' }}\n {{- '\"parameters\": ' }}\n {{- tool_call.arguments | tojson }}\n {{- \"}\" }}\n {%- endif %}\n {%- if builtin_tools is defined %}\n {#- This means we're in ipython mode #}\n {{- \"<|eom_id|>\" }}\n {%- else %}\n {{- \"<|eot_id|>\" }}\n {%- endif %}\n {%- elif message.role == \"tool\" or message.role == \"ipython\" %}\n {{- \"<|start_header_id|>ipython<|end_header_id|>\\n\\n\" }}\n {%- if message.content is mapping or message.content is iterable %}\n {{- message.content | tojson }}\n {%- else %}\n {{- message.content }}\n {%- endif %}\n {{- \"<|eot_id|>\" }}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|start_header_id|>assistant<|end_header_id|>\\n\\n' }}\n{%- endif %}\n", + "clean_up_tokenization_spaces": true, + "eos_token": "<|eot_id|>", + "legacy": false, + "model_input_names": [ + "input_ids", + "attention_mask" + ], + "model_max_length": 131072, + "tokenizer_class": "PreTrainedTokenizerFast" +}