n1ck-guo's picture
upload auto_round format
1bfb497
{
"_name_or_path": "/models/falcon3-3B",
"activation": "swiglu",
"architectures": [
"LlamaForCausalLM"
],
"attention_bias": false,
"attention_dropout": 0.0,
"bos_token_id": 11,
"eos_token_id": 11,
"head_dim": 256,
"hidden_act": "silu",
"hidden_size": 3072,
"initializer_range": 0.02,
"intermediate_size": 9216,
"max_position_embeddings": 4096,
"mlp_bias": false,
"model_type": "llama",
"num_attention_heads": 12,
"num_hidden_layers": 22,
"num_key_value_heads": 4,
"parallel_attn": false,
"pretraining_tp": 1,
"quantization_config": {
"amp": true,
"autoround_version": "0.4.2",
"backend": "auto_round:gptq:exllamav2",
"batch_size": 8,
"bits": 4,
"data_type": "int",
"dataset": "NeelNanda/pile-10k",
"enable_minmax_tuning": true,
"enable_norm_bias_tuning": false,
"enable_quanted_input": true,
"gradient_accumulate_steps": 1,
"group_size": 128,
"iters": 2000,
"low_gpu_mem_usage": false,
"lr": 0.0005,
"minmax_lr": 0.0005,
"nsamples": 512,
"quant_method": "intel/auto-round",
"scale_dtype": "torch.float16",
"seqlen": 2048,
"sym": true,
"to_quant_block_names": null
},
"rms_norm_eps": 1e-06,
"rope_scaling": null,
"rope_theta": 1000042,
"tie_word_embeddings": false,
"torch_dtype": "float16",
"transformers_version": "4.47.0",
"use_cache": true,
"vocab_size": 131072
}