config.json

{
  "architectures": [
    "Eagle3Speculator"
  ],
  "speculators_model_type": "eagle3",
  "speculators_version": "0.1.0.dev42",
  "draft_vocab_size": 32000,
  "norm_before_residual": false,
  "target_hidden_size": null,
  "eagle_aux_hidden_state_layer_ids": [
    1,
    45,
    89
  ],
  "transformer_layer_config": {
    "model_type": "llama",
    "vocab_size": 151936,
    "hidden_size": 4096,
    "intermediate_size": 12288,
    "num_hidden_layers": 1,
    "num_attention_heads": 64,
    "num_key_value_heads": 4,
    "head_dim": 128,
    "hidden_act": "silu",
    "max_position_embeddings": 262144,
    "initializer_range": 0.02,
    "rms_norm_eps": 1e-05,
    "pretraining_tp": 1,
    "use_cache": true,
    "rope_theta": 500000.0,
    "rope_scaling": {
      "factor": 8.0,
      "high_freq_factor": 4.0,
      "low_freq_factor": 1.0,
      "original_max_position_embeddings": 8192,
      "rope_type": "llama3"
    },
    "attention_bias": false,
    "attention_dropout": 0.0,
    "mlp_bias": false,
    "tie_word_embeddings": false
  },
  "speculators_config": {
    "algorithm": "eagle3",
    "default_proposal_method": "greedy",
    "proposal_methods": [
      {
        "proposal_type": "greedy",
        "speculative_tokens": 3,
        "accept_tolerance": 0.0
      }
    ],
    "verifier": {
      "name_or_path": "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8",
      "architectures": [
        "Qwen3MoeForCausalLM"
      ]
    }
  },
  "torch_dtype": "bfloat16",
  "_comment": "Eagle3 head based on Llama3 architecture targeting Qwen3 verifier",
  "_conversion_notes": {
    "source": "nm-testing/Mockup-qwen235-eagle3-fp16",
    "architecture_notes": "Eagle3 head uses Llama3 rope_type, targets Qwen3 verifier",
    "vocabulary_notes": "Draft vocab size 32000, target vocab size 151936",
    "auxiliary_layers": "Uses hidden states from verifier layers 3, 47, 91",
    "implementation_note": "May require Eagle3Speculator extensions for aux hidden states"
  }
}