Upload config.json with huggingface_hub
Browse files- config.json +85 -0
config.json
ADDED
|
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "/checkpoint/swiftkv/llama-swiftkv-405b-20241106-034807-layer63-group1/global_step-2118/",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"LlamaSwiftKVForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 128000,
|
| 9 |
+
"eos_token_id": [
|
| 10 |
+
128001,
|
| 11 |
+
128008,
|
| 12 |
+
128009
|
| 13 |
+
],
|
| 14 |
+
"head_dim": 128,
|
| 15 |
+
"hidden_act": "silu",
|
| 16 |
+
"hidden_size": 16384,
|
| 17 |
+
"initializer_range": 0.02,
|
| 18 |
+
"intermediate_size": 53248,
|
| 19 |
+
"key_value_group_size": 1,
|
| 20 |
+
"max_position_embeddings": 131072,
|
| 21 |
+
"mlp_bias": false,
|
| 22 |
+
"model_type": "llama_swiftkv",
|
| 23 |
+
"num_attention_heads": 128,
|
| 24 |
+
"num_hidden_layers": 126,
|
| 25 |
+
"num_key_value_heads": 8,
|
| 26 |
+
"num_key_value_layers": 63,
|
| 27 |
+
"pretraining_tp": 1,
|
| 28 |
+
"quantization_config": {
|
| 29 |
+
"config_groups": {
|
| 30 |
+
"group_0": {
|
| 31 |
+
"input_activations": {
|
| 32 |
+
"actorder": null,
|
| 33 |
+
"block_structure": null,
|
| 34 |
+
"dynamic": true,
|
| 35 |
+
"group_size": null,
|
| 36 |
+
"num_bits": 8,
|
| 37 |
+
"observer": null,
|
| 38 |
+
"observer_kwargs": {},
|
| 39 |
+
"strategy": "token",
|
| 40 |
+
"symmetric": true,
|
| 41 |
+
"type": "float"
|
| 42 |
+
},
|
| 43 |
+
"output_activations": null,
|
| 44 |
+
"targets": [
|
| 45 |
+
"Linear"
|
| 46 |
+
],
|
| 47 |
+
"weights": {
|
| 48 |
+
"actorder": null,
|
| 49 |
+
"block_structure": null,
|
| 50 |
+
"dynamic": false,
|
| 51 |
+
"group_size": null,
|
| 52 |
+
"num_bits": 8,
|
| 53 |
+
"observer": "minmax",
|
| 54 |
+
"observer_kwargs": {},
|
| 55 |
+
"strategy": "channel",
|
| 56 |
+
"symmetric": true,
|
| 57 |
+
"type": "float"
|
| 58 |
+
}
|
| 59 |
+
}
|
| 60 |
+
},
|
| 61 |
+
"format": "float-quantized",
|
| 62 |
+
"global_compression_ratio": 1.512666303241781,
|
| 63 |
+
"ignore": [
|
| 64 |
+
"lm_head"
|
| 65 |
+
],
|
| 66 |
+
"kv_cache_scheme": null,
|
| 67 |
+
"quant_method": "compressed-tensors",
|
| 68 |
+
"quantization_status": "compressed"
|
| 69 |
+
},
|
| 70 |
+
"rms_norm_eps": 1e-05,
|
| 71 |
+
"rope_scaling": {
|
| 72 |
+
"factor": 8.0,
|
| 73 |
+
"high_freq_factor": 4.0,
|
| 74 |
+
"low_freq_factor": 1.0,
|
| 75 |
+
"original_max_position_embeddings": 8192,
|
| 76 |
+
"rope_type": "llama3"
|
| 77 |
+
},
|
| 78 |
+
"rope_theta": 500000.0,
|
| 79 |
+
"swiftkv": true,
|
| 80 |
+
"tie_word_embeddings": false,
|
| 81 |
+
"torch_dtype": "bfloat16",
|
| 82 |
+
"transformers_version": "4.46.2",
|
| 83 |
+
"use_cache": true,
|
| 84 |
+
"vocab_size": 128256
|
| 85 |
+
}
|