{
  "architectures": [
    "Qwen3ForCausalLM"
  ],
  "model_type": "qwen3",
  "quantization_config": {
    "quant_method": "rocmfpx",
    "quant_type": "ROCmFP4_FAST",
    "bits_per_weight": 4.26,
    "block_size": 32,
    "cooperative_matrix_target": "gfx1151",
    "mtp_heads": 1,
    "turboquant_kv": true
  },
  "vocab_size": 152064,
  "hidden_size": 5120,
  "intermediate_size": 27648,
  "num_hidden_layers": 64,
  "num_attention_heads": 40,
  "num_key_value_heads": 8,
  "max_position_embeddings": 262144,
  "rope_scaling": {
    "type": "yarn",
    "factor": 4.0,
    "original_max_position_embeddings": 32768
  },
  "torch_dtype": "bfloat16",
  "transformers_version": "4.45.0"
}