{
  "version": "0.1.0",
  "model_type": "phi3",
  "quantization": "q4f16_1",
  "model_config": {
    "model_type": "phi3",
    "hidden_size": 3072,
    "vocab_size": 32064,
    "num_hidden_layers": 32,
    "num_attention_heads": 32,
    "intermediate_size": 8192,
    "rms_norm_eps": 1e-05,
    "num_key_value_heads": 32,
    "position_embedding_base": 10000.0,
    "context_window_size": 131072,
    "prefill_chunk_size": 2048,
    "head_dim": 96,
    "tensor_parallel_shards": 1,
    "max_batch_size": 80
  },
  "vocab_size": 32064,
  "context_window_size": 131072,
  "sliding_window_size": -1,
  "prefill_chunk_size": 2048,
  "attention_sink_size": -1,
  "tensor_parallel_shards": 1,
  "temperature": 1.0,
  "presence_penalty": 0.0,
  "frequency_penalty": 0.0,
  "repetition_penalty": 1.0,
  "top_p": 1.0,
  "tokenizer_files": [
    "tokenizer.model",
    "tokenizer.json",
    "added_tokens.json",
    "tokenizer_config.json"
  ],
  "tokenizer_info": {
    "token_postproc_method": "byte_fallback",
    "prepend_space_in_encode": true,
    "strip_space_in_decode": true
  },
  "conv_template": {
    "name": "phi-3",
    "system_template": "<|system|>\n{system_message}",
    "system_message": "You are a helpful digital assistant. Please provide safe, ethical and accurate information to the user.",
    "system_prefix_token_ids": [
      1
    ],
    "add_role_after_system_message": true,
    "roles": {
      "user": "<|user|>",
      "assistant": "<|assistant|>"
    },
    "role_templates": {
      "user": "{user_message}",
      "assistant": "{assistant_message}",
      "tool": "{tool_message}"
    },
    "messages": [],
    "seps": [
      "<|end|>\n"
    ],
    "role_content_sep": "\n",
    "role_empty_sep": "\n",
    "stop_str": [
      "<|endoftext|>"
    ],
    "stop_token_ids": [
      32000,
      32001,
      32007
    ],
    "function_string": "",
    "use_function_calling": false
  },
  "pad_token_id": 32000,
  "bos_token_id": 1,
  "eos_token_id": [
    32000,
    32001,
    32007
  ],
  "mean_gen_len": 128,
  "max_gen_len": 512,
  "shift_fill_factor": 0.3
}