config.json

{
  "architectures": [
    "Qwen3Model"
  ],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "dtype": "bfloat16",
  "head_dim": 128,
  "hidden_size": 4096,
  "initializer_range": 0.02,
  "max_position_embeddings": 32768,
  "model_type": "qwen3",
  "num_attention_heads": 32,
  "num_hidden_layers": 32,
  "num_key_value_heads": 32,
  "rms_norm_eps": 1e-06,
  "transformers_version": "4.56.0",
  "vocab_size": 151936
}