k-l-lambda commited on
Commit
29a126e
·
verified ·
1 Parent(s): 21b55f6

config: Eagle3DeepseekV2ForCausalLM + MLA params, aux_layers=[2,30,58]

Browse files
Files changed (1) hide show
  1. config.json +50 -70
config.json CHANGED
@@ -1,76 +1,56 @@
1
  {
2
  "architectures": [
3
- "Eagle3DraftModel"
4
  ],
5
- "auto_map": {
6
- "": "config.Eagle3SpeculatorConfig"
7
- },
8
- "base_model_ep_plan": null,
9
- "draft_vocab_size": 32000,
10
- "dtype": "float32",
11
- "eagle_aux_hidden_state_layer_ids": null,
12
- "embed_requires_grad": false,
13
- "has_no_defaults_at_init": false,
14
- "norm_before_residual": true,
15
- "speculators_config": {
16
- "algorithm": "eagle3",
17
- "default_proposal_method": "greedy",
18
- "proposal_methods": [
19
- {
20
- "accept_tolerance": 0.0,
21
- "proposal_type": "greedy",
22
- "speculative_tokens": 3,
23
- "verifier_accept_k": 1
24
- }
25
  ],
26
- "verifier": {
27
- "architectures": [],
28
- "name_or_path": "/data/.cache_claude/huggingface/hub/models--moonshotai--Kimi-K2.5/snapshots/54383e83fa343a1331754112fb9e3410c55efa2f"
29
- }
30
  },
31
- "speculators_model_type": "eagle3",
32
- "speculators_version": "0.5.0.dev1",
33
- "target_hidden_size": null,
34
- "transformer_layer_config": {
35
- "attention_bias": false,
36
- "attention_dropout": 0.0,
37
- "aux_loss_alpha": 0.001,
38
- "ep_size": 1,
39
- "first_k_dense_replace": 3,
40
- "head_dim": null,
41
- "hidden_act": "silu",
42
- "hidden_size": 7168,
43
- "initializer_range": 0.02,
44
- "intermediate_size": 18432,
45
- "kv_lora_rank": 512,
46
- "max_position_embeddings": 262144,
47
- "model_type": "deepseek_v3",
48
- "moe_intermediate_size": 2048,
49
- "moe_layer_freq": 1,
50
- "n_group": 8,
51
- "n_routed_experts": 256,
52
- "n_shared_experts": 1,
53
- "norm_topk_prob": true,
54
- "num_attention_heads": 64,
55
- "num_experts_per_tok": 8,
56
- "num_hidden_layers": 1,
57
- "num_key_value_heads": 64,
58
- "num_nextn_predict_layers": 1,
59
- "pretraining_tp": 1,
60
- "q_lora_rank": 1536,
61
- "qk_nope_head_dim": 128,
62
- "qk_rope_head_dim": 64,
63
- "rms_norm_eps": 1e-05,
64
- "rope_scaling": null,
65
- "rope_theta": 10000.0,
66
- "routed_scaling_factor": 2.5,
67
- "scoring_func": "sigmoid",
68
- "seq_aux": true,
69
- "topk_group": 4,
70
- "topk_method": "noaux_tc",
71
- "use_cache": true,
72
- "v_head_dim": 128,
73
- "vocab_size": 163840
74
  },
75
- "transformers_version": "4.57.6"
76
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  {
2
  "architectures": [
3
+ "Eagle3DeepseekV2ForCausalLM"
4
  ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 163584,
8
+ "draft_vocab_size": 163840,
9
+ "dtype": "bfloat16",
10
+ "eagle_config": {
11
+ "eagle_aux_hidden_state_layer_ids": [
12
+ 2,
13
+ 30,
14
+ 58
 
 
 
 
 
 
 
 
 
 
15
  ],
16
+ "use_aux_hidden_state": true
 
 
 
17
  },
18
+ "eos_token_id": 163585,
19
+ "head_dim": 128,
20
+ "hidden_act": "silu",
21
+ "hidden_size": 7168,
22
+ "initializer_range": 0.02,
23
+ "intermediate_size": 18432,
24
+ "max_position_embeddings": 262144,
25
+ "max_window_layers": 48,
26
+ "mlp_bias": false,
27
+ "model_type": "kimi_k2",
28
+ "num_attention_heads": 64,
29
+ "num_hidden_layers": 1,
30
+ "num_key_value_heads": 64,
31
+ "pretraining_tp": 1,
32
+ "rms_norm_eps": 1e-05,
33
+ "rope_scaling": {
34
+ "beta_fast": 32.0,
35
+ "beta_slow": 1.0,
36
+ "factor": 64.0,
37
+ "mscale": 1.0,
38
+ "mscale_all_dim": 1.0,
39
+ "original_max_position_embeddings": 4096,
40
+ "type": "yarn"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
41
  },
42
+ "rope_theta": 50000.0,
43
+ "sliding_window": null,
44
+ "tie_word_embeddings": false,
45
+ "transformers_version": "4.57.1",
46
+ "use_cache": true,
47
+ "use_sliding_window": false,
48
+ "vocab_size": 163840,
49
+ "qk_nope_head_dim": 128,
50
+ "qk_rope_head_dim": 64,
51
+ "q_lora_rank": 1536,
52
+ "kv_lora_rank": 512,
53
+ "v_head_dim": 128,
54
+ "first_k_dense_replace": 1,
55
+ "moe_layer_freq": 1
56
+ }