config.json

{
  "_name_or_path": "DeepSeek-R1-Distill-Qwen-7B",
  "architectures": [
    "Qwen2ForCausalLM"
  ],
  "asym": false,
  "attention_dropout": 0.0,
  "bigdl_transformers_low_bit": "sym_int4_rtn",
  "bos_token_id": 151643,
  "const_parameter": true,
  "cos_sin_input": true,
  "embedding_post": true,
  "eos_token_id": 151643,
  "fused_layers": 2,
  "group_size": 0,
  "head_dim": 128,
  "hidden_act": "silu",
  "hidden_size": 3584,
  "initializer_range": 0.02,
  "intermediate_size": 18944,
  "kv_len": 1023,
  "lm_head_low_bit": "sym_int4_rtn",
  "max_position_embeddings": 131072,
  "max_prompt_len": 512,
  "max_window_layers": 28,
  "mixed_precision": false,
  "model_type": "qwen2",
  "n_splits_down_proj": 2,
  "n_splits_linear": 1,
  "num_attention_heads": 28,
  "num_head": 28,
  "num_hidden_layers": 28,
  "num_key_value_heads": 4,
  "optimize_model": true,
  "qkv_bias": true,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 10000,
  "sliding_window": null,
  "tie_word_embeddings": false,
  "torch_dtype": "float16",
  "transformers_version": "4.45.0",
  "transpose_value_cache": true,
  "use_cache": true,
  "use_mrope": false,
  "use_prefill_sdp": false,
  "use_sliding_window": false,
  "vocab_size": 152064,
  "weight_idx": 8,
  "weight_num": 7
}