Add q004 EAGLE3 eagle3-draft-chat-think weights

eagle3-draft-chat-think/config.json

@@ -0,0 +1,68 @@
+{
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 151643,
+  "draft_vocab_size": 32768,
+  "eagle_config": {
+    "base_model": "/home/project-a/Efficient-Qwen-Competition/models/qwen35-4b-awq-text-only-packed",
+    "stateless_fusion": true,
+    "use_aux_hidden_state": false
+  },
+  "eos_token_id": 248044,
+  "head_dim": 256,
+  "hidden_act": "silu",
+  "hidden_size": 2560,
+  "intermediate_size": 9216,
+  "max_position_embeddings": 262144,
+  "model_type": "llama",
+  "num_attention_heads": 10,
+  "num_hidden_layers": 0,
+  "num_key_value_heads": 2,
+  "pad_token_id": 248044,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 1000000.0,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "training": {
+    "absolute_positions": true,
+    "adam_beta1": 0.9,
+    "adam_beta2": 0.95,
+    "batch_size": 1,
+    "cache_aware_draft_steps": true,
+    "dense_target_hidden_soft_loss": true,
+    "dense_target_temperature": 0.65,
+    "draft_head_dim": 0,
+    "draft_num_attention_heads": 0,
+    "draft_num_hidden_layers": 0,
+    "draft_num_key_value_heads": 0,
+    "draft_rope_theta": 0.0,
+    "draft_step_decay": 0.9,
+    "draft_steps": 4,
+    "draft_vocab_jsonl": "/home/project-a/efficient-qwen/tmp/repro_awq_think_4k_20260520/features/chat_think/metadata.jsonl",
+    "draft_vocab_size": 32768,
+    "drop_longer_than_max_seq_len": false,
+    "feature_glob": "/home/project-a/efficient-qwen/tmp/repro_awq_think_4k_20260520/features/chat_think/features_*.pt",
+    "gradient_clipping": 0.5,
+    "hard_target_weight": 0.75,
+    "label_type": "onpolicy_teacher_dense_draft_vocab",
+    "learning_rate": 6e-06,
+    "max_seq_len": 4096,
+    "nemo_draft_loss": false,
+    "no_aux_hidden_state": true,
+    "num_examples": 4096,
+    "official_soft_loss": false,
+    "parallel_drafting_anchors_per_example": 1,
+    "parallel_drafting_mask_hidden": false,
+    "parallel_drafting_objective": false,
+    "parallel_drafting_ptd_token_id": -1,
+    "save_every": 3000,
+    "shift_positions_per_draft_step": false,
+    "soft_target_weight": 1.0,
+    "steps": 12000,
+    "target_hidden_sidecar_map": "",
+    "warmup_steps": 500
+  },
+  "use_cache": true,
+  "vocab_size": 248320
+}

eagle3-draft-chat-think/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0aceb4f06084e9b72cb908885c8b12e91cfb508ccc3e6baf4a9c22c8fd21a1bf
+size 194264600