config.json

{
  "architectures": [
    "DFlashDraftModel"
  ],
  "model_type": "qwen3",
  "block_size": 8,
  "dflash_config": {
    "mask_token_id": 163838,
    "target_layer_ids": [
      1,
      12,
      24,
      35,
      47,
      58
    ]
  },
  "hidden_size": 7168,
  "num_hidden_layers": 6,
  "num_attention_heads": 64,
  "num_key_value_heads": 8,
  "head_dim": 128,
  "intermediate_size": 18432,
  "hidden_act": "silu",
  "rms_norm_eps": 1e-06,
  "vocab_size": 163840,
  "max_position_embeddings": 262144,
  "initializer_range": 0.02,
  "attention_bias": false,
  "attention_dropout": 0.0,
  "rope_theta": 50000.0,
  "rope_scaling": {
    "beta_fast": 32.0,
    "beta_slow": 1.0,
    "factor": 16.0,
    "mscale": 1.0,
    "mscale_all_dim": 1.0,
    "original_max_position_embeddings": 4096,
    "rope_type": "yarn",
    "type": "yarn"
  },
  "tie_word_embeddings": false,
  "torch_dtype": "bfloat16",
  "num_target_layers": 61,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ]
}