{
  "version": "0.1.0",
  "model_type": "asvd_phi",
  "quantization": "q0f16",
  "model_config": {
    "vocab_size": 51200,
    "hidden_size": 2560,
    "intermediate_size": 10240,
    "num_hidden_layers": 32,
    "num_attention_heads": 32,
    "layer_norm_eps": 1e-05,
    "position_embedding_base": 10000.0,
    "partial_rotary_factor": 0.4,
    "num_key_value_heads": 32,
    "context_window_size": 128,
    "prefill_chunk_size": 128,
    "head_dim": 80,
    "truncation_ranks": {
      "model.layers.0.mlp.fc1": [
        1475,
        163
      ],
      "model.layers.0.mlp.fc2": [
        1290,
        143
      ],
      "model.layers.0.self_attn.dense": [
        922,
        102
      ],
      "model.layers.0.self_attn.k_proj": [
        922,
        102
      ],
      "model.layers.0.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.1.mlp.fc1": [
        1475,
        163
      ],
      "model.layers.1.self_attn.dense": [
        807,
        89
      ],
      "model.layers.1.self_attn.k_proj": [
        692,
        76
      ],
      "model.layers.1.self_attn.v_proj": [
        576,
        64
      ],
      "model.layers.10.self_attn.q_proj": [
        692,
        76
      ],
      "model.layers.11.self_attn.dense": [
        461,
        51
      ],
      "model.layers.11.self_attn.v_proj": [
        692,
        76
      ],
      "model.layers.12.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.12.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.12.self_attn.v_proj": [
        922,
        102
      ],
      "model.layers.13.self_attn.dense": [
        807,
        89
      ],
      "model.layers.13.self_attn.q_proj": [
        576,
        64
      ],
      "model.layers.14.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.14.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.15.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.15.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.16.self_attn.dense": [
        922,
        102
      ],
      "model.layers.16.self_attn.k_proj": [
        922,
        102
      ],
      "model.layers.17.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.17.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.18.self_attn.dense": [
        922,
        102
      ],
      "model.layers.19.self_attn.k_proj": [
        922,
        102
      ],
      "model.layers.19.self_attn.v_proj": [
        922,
        102
      ],
      "model.layers.2.mlp.fc1": [
        1475,
        163
      ],
      "model.layers.2.self_attn.dense": [
        807,
        89
      ],
      "model.layers.2.self_attn.k_proj": [
        1037,
        115
      ],
      "model.layers.2.self_attn.q_proj": [
        461,
        51
      ],
      "model.layers.20.self_attn.dense": [
        807,
        89
      ],
      "model.layers.20.self_attn.q_proj": [
        1037,
        115
      ],
      "model.layers.21.self_attn.dense": [
        692,
        76
      ],
      "model.layers.21.self_attn.q_proj": [
        1037,
        115
      ],
      "model.layers.22.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.22.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.23.self_attn.dense": [
        922,
        102
      ],
      "model.layers.23.self_attn.q_proj": [
        922,
        102
      ],
      "model.layers.23.self_attn.v_proj": [
        692,
        76
      ],
      "model.layers.24.self_attn.dense": [
        461,
        51
      ],
      "model.layers.24.self_attn.q_proj": [
        922,
        102
      ],
      "model.layers.25.self_attn.k_proj": [
        576,
        64
      ],
      "model.layers.26.self_attn.k_proj": [
        922,
        102
      ],
      "model.layers.26.self_attn.q_proj": [
        1037,
        115
      ],
      "model.layers.27.self_attn.dense": [
        692,
        76
      ],
      "model.layers.27.self_attn.k_proj": [
        807,
        89
      ],
      "model.layers.28.self_attn.dense": [
        576,
        64
      ],
      "model.layers.28.self_attn.q_proj": [
        922,
        102
      ],
      "model.layers.29.self_attn.q_proj": [
        922,
        102
      ],
      "model.layers.29.self_attn.v_proj": [
        807,
        89
      ],
      "model.layers.3.self_attn.dense": [
        922,
        102
      ],
      "model.layers.3.self_attn.k_proj": [
        1037,
        115
      ],
      "model.layers.30.self_attn.dense": [
        922,
        102
      ],
      "model.layers.30.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.31.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.31.self_attn.q_proj": [
        1037,
        115
      ],
      "model.layers.4.self_attn.dense": [
        807,
        89
      ],
      "model.layers.4.self_attn.k_proj": [
        1037,
        115
      ],
      "model.layers.4.self_attn.q_proj": [
        692,
        76
      ],
      "model.layers.5.self_attn.k_proj": [
        1037,
        115
      ],
      "model.layers.5.self_attn.q_proj": [
        807,
        89
      ],
      "model.layers.6.self_attn.dense": [
        807,
        89
      ],
      "model.layers.6.self_attn.q_proj": [
        922,
        102
      ],
      "model.layers.7.self_attn.q_proj": [
        692,
        76
      ],
      "model.layers.8.self_attn.dense": [
        1037,
        115
      ],
      "model.layers.8.self_attn.k_proj": [
        692,
        76
      ],
      "model.layers.8.self_attn.q_proj": [
        1037,
        115
      ],
      "model.layers.9.self_attn.dense": [
        692,
        76
      ],
      "model.layers.9.self_attn.q_proj": [
        576,
        64
      ]
    },
    "tensor_parallel_shards": 1,
    "max_batch_size": 1
  },
  "vocab_size": 51200,
  "context_window_size": 2048,
  "sliding_window_size": -1,
  "prefill_chunk_size": 2048,
  "attention_sink_size": -1,
  "tensor_parallel_shards": 1,
  "pipeline_parallel_stages": 1,
  "temperature": 1.0,
  "presence_penalty": 0.0,
  "frequency_penalty": 0.0,
  "repetition_penalty": 1.0,
  "top_p": 1.0,
  "tokenizer_files": [
    "tokenizer.json",
    "vocab.json",
    "merges.txt",
    "added_tokens.json",
    "tokenizer_config.json"
  ],
  "tokenizer_info": {
    "token_postproc_method": "byte_level",
    "prepend_space_in_encode": false,
    "strip_space_in_decode": false
  },
  "conv_template": {
    "name": "LM",
    "system_template": "{system_message}",
    "system_message": "",
    "system_prefix_token_ids": [
      1
    ],
    "add_role_after_system_message": true,
    "roles": {
      "user": "",
      "assistant": ""
    },
    "role_templates": {
      "user": "{user_message}",
      "assistant": "{assistant_message}",
      "tool": "{tool_message}"
    },
    "messages": [],
    "seps": [
      ""
    ],
    "role_content_sep": "",
    "role_empty_sep": "",
    "stop_str": [],
    "stop_token_ids": [
      2
    ],
    "function_string": "",
    "use_function_calling": false
  },
  "pad_token_id": 0,
  "bos_token_id": 50256,
  "eos_token_id": 50256
}