A.X-4.0-AWQ-tp1 / config.json
dev7halo's picture
Upload Qwen2ForCausalLM
d20f957 verified
{
"architectures": [
"Qwen2ForCausalLM"
],
"attention_dropout": 0.0,
"bos_token_id": 0,
"eos_token_id": 0,
"hidden_act": "silu",
"hidden_size": 8192,
"initializer_range": 0.02,
"intermediate_size": 29568,
"max_position_embeddings": 131072,
"max_window_layers": 28,
"model_type": "qwen2",
"num_attention_heads": 64,
"num_hidden_layers": 80,
"num_key_value_heads": 8,
"quantization_config": {
"backend": "autoawq",
"bits": 4,
"do_fuse": false,
"exllama_config": null,
"fuse_max_seq_len": null,
"group_size": 128,
"modules_to_fuse": null,
"modules_to_not_convert": null,
"quant_method": "awq",
"version": "gemm",
"zero_point": true
},
"rms_norm_eps": 1e-05,
"rope_scaling": {
"factor": 2.0,
"original_max_position_embeddings": 131072,
"rope_type": "yarn",
"type": "yarn"
},
"rope_theta": 5000000,
"sliding_window": null,
"tie_word_embeddings": false,
"torch_dtype": "float16",
"transformers_version": "4.52.3",
"use_cache": false,
"use_sliding_window": false,
"vocab_size": 102400
}