reference/code/SpecForge/configs/lfm2.5-1.2b-instruct-dflash.json

{
  "architectures": ["DFlashDraftModel"],
  "attention_bias": false,
  "attention_dropout": 0.0,
  "auto_map": {
    "AutoModel": "dflash.DFlashDraftModel"
  },
  "block_size": 16,
  "bos_token_id": 1,
  "dflash_config": {
    "mask_token_id": 16,
    "target_layer_ids": [1, 4, 7, 10, 13]
  },
  "dtype": "bfloat16",
  "eos_token_id": 7,
  "head_dim": 64,
  "hidden_act": "silu",
  "hidden_size": 2048,
  "initializer_range": 0.02,
  "intermediate_size": 6144,
  "layer_types": [
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention",
    "full_attention"
  ],
  "max_position_embeddings": 128000,
  "model_type": "qwen3",
  "num_attention_heads": 32,
  "num_hidden_layers": 8,
  "num_key_value_heads": 8,
  "num_target_layers": 16,
  "rms_norm_eps": 1e-05,
  "rope_theta": 1000000.0,
  "tie_word_embeddings": true,
  "use_cache": true,
  "vocab_size": 65536
}