config.json

{
  "_name_or_path": "/data/llm/grok-1",
  "architectures": [
    "Grok1ModelForCausalLM"
  ],
  "attn_output_multiplier": 0.08838834764831845,
  "auto_map": {
    "AutoConfig": "configuration_grok1.Grok1Config",
    "AutoModel": "modeling_grok1.Grok1Model",
    "AutoModelForCausalLM": "modeling_grok1.Grok1ModelForCausalLM"
  },
  "bos_token_id": 1,
  "embedding_multiplier_scale": 78.38367176906169,
  "eos_token_id": 2,
  "hidden_size": 6144,
  "intermediate_size": 32768,
  "max_attn_value": 30.0,
  "max_position_embeddings": 8192,
  "model_type": "grok-1",
  "num_attention_heads": 48,
  "num_experts": 8,
  "num_experts_per_tok": 2,
  "num_hidden_layers": 64,
  "num_key_value_heads": 8,
  "output_multiplier_scale": 0.5773502691896257,
  "output_router_logits": false,
  "pad_token_id": 0,
  "quantization_config": {
    "activation_scheme": "static",
    "ignored_layers": [
      "lm_head",
      "*.gate"
    ],
    "kv_cache_scheme": "static",
    "quant_method": "fp8"
  },
  "rms_norm_eps": 1e-05,
  "router_aux_loss_coef": 0.001,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.45.0.dev0",
  "use_cache": true,
  "vocab_size": 131072
}