config.json

{
  "architectures": [
    "DFlashDraftModel"
  ],
  "auto_map": {
    "": "config.DFlashSpeculatorConfig"
  },
  "aux_hidden_state_layer_ids": [
    2,
    16,
    29
  ],
  "block_size": 16,
  "draft_vocab_size": 32000,
  "dtype": "bfloat16",
  "mask_token_id": 248077,
  "max_anchors": 256,
  "sliding_window_non_causal": false,
  "speculators_config": {
    "algorithm": "dflash",
    "default_proposal_method": "greedy",
    "proposal_methods": [
      {
        "accept_tolerance": 0.0,
        "proposal_type": "greedy",
        "speculative_tokens": 15,
        "verifier_accept_k": 1
      }
    ],
    "verifier": {
      "architectures": [],
      "name_or_path": "RedHatAI/Qwen3.5-4B-quantized.w4a16"
    }
  },
  "speculators_model_type": "dflash",
  "speculators_version": "0.5.0.dev166",
  "target_hidden_size": null,
  "tie_word_embeddings": false,
  "transformer_layer_config": {
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": 1,
    "eos_token_id": 2,
    "head_dim": 256,
    "hidden_act": "silu",
    "hidden_size": 2560,
    "initializer_range": 0.02,
    "intermediate_size": 9216,
    "layer_types": [
      "full_attention",
      "full_attention"
    ],
    "max_position_embeddings": 262144,
    "mlp_bias": false,
    "model_type": "llama",
    "num_attention_heads": 16,
    "num_hidden_layers": 2,
    "num_key_value_heads": 4,
    "pad_token_id": null,
    "pretraining_tp": 1,
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
      "mrope_interleaved": true,
      "mrope_section": [
        11,
        11,
        10
      ],
      "partial_rotary_factor": 0.25,
      "rope_theta": 10000000,
      "rope_type": "default"
    },
    "sliding_window": 2048,
    "tie_word_embeddings": false,
    "use_cache": true,
    "vocab_size": 248320
  },
  "transformers_version": "5.6.2"
}