Hugging Face's logo

keras
/
qwen3_coder_instruct_30b_a3b_en

Text Generation
KerasHub
Model card Files
xet
 Community
prasadsachin commited on
Commit
2f57005
·
verified ·
1 Parent(s): 94a0db3

Upload folder using huggingface_hub

Browse files
Files changed (22) hide show
  1. README.md +33 -0
  2. assets/tokenizer/merges.txt +0 -0
  3. assets/tokenizer/vocabulary.json +0 -0
  4. config.json +37 -0
  5. metadata.json +9 -0
  6. model.weights.json +587 -0
  7. model_00000.weights.h5 +3 -0
  8. model_00001.weights.h5 +3 -0
  9. model_00002.weights.h5 +3 -0
  10. model_00003.weights.h5 +3 -0
  11. model_00004.weights.h5 +3 -0
  12. model_00005.weights.h5 +3 -0
  13. model_00006.weights.h5 +3 -0
  14. model_00007.weights.h5 +3 -0
  15. model_00008.weights.h5 +3 -0
  16. model_00009.weights.h5 +3 -0
  17. model_00010.weights.h5 +3 -0
  18. model_00011.weights.h5 +3 -0
  19. model_00012.weights.h5 +3 -0
  20. preprocessor.json +69 -0
  21. task.json +114 -0
  22. tokenizer.json +48 -0
README.md ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: keras-hub
3
+ pipeline_tag: text-generation
4
+ ---
5
+ This is a [`Qwen3Moe` model](https://keras.io/api/keras_hub/models/qwen3_moe) uploaded using the KerasHub library and can be used with JAX, TensorFlow, and PyTorch backends.
6
+ This model is related to a `CausalLM` task.
7
+
8
+ Model config:
9
+ * **name:** qwen3_moe_backbone
10
+ * **trainable:** True
11
+ * **dtype:** {'module': 'keras', 'class_name': 'DTypePolicy', 'config': {'name': 'float32'}, 'registered_name': None}
12
+ * **vocabulary_size:** 151936
13
+ * **num_layers:** 48
14
+ * **num_query_heads:** 32
15
+ * **head_dim:** 128
16
+ * **hidden_dim:** 2048
17
+ * **intermediate_dim:** 6144
18
+ * **moe_intermediate_dim:** 768
19
+ * **rope_max_wavelength:** 10000000
20
+ * **num_key_value_heads:** 4
21
+ * **rope_scaling_factor:** 1.0
22
+ * **layer_norm_epsilon:** 1e-06
23
+ * **dropout:** 0
24
+ * **tie_word_embeddings:** False
25
+ * **sliding_window_size:** None
26
+ * **num_experts:** 128
27
+ * **top_k:** 8
28
+ * **norm_top_k_prob:** True
29
+ * **decoder_sparse_step:** 1
30
+ * **mlp_only_layers:** []
31
+ * **router_aux_loss_coefficient:** 0.001
32
+
33
+ This model card has been generated automatically and should be completed by the model author. See [Model Cards documentation](https://huggingface.co/docs/hub/model-cards) for more information.
assets/tokenizer/merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
assets/tokenizer/vocabulary.json ADDED
The diff for this file is too large to render. See raw diff
 
config.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_backbone",
3
+ "class_name": "Qwen3MoeBackbone",
4
+ "config": {
5
+ "name": "qwen3_moe_backbone",
6
+ "trainable": true,
7
+ "dtype": {
8
+ "module": "keras",
9
+ "class_name": "DTypePolicy",
10
+ "config": {
11
+ "name": "float32"
12
+ },
13
+ "registered_name": null
14
+ },
15
+ "vocabulary_size": 151936,
16
+ "num_layers": 48,
17
+ "num_query_heads": 32,
18
+ "head_dim": 128,
19
+ "hidden_dim": 2048,
20
+ "intermediate_dim": 6144,
21
+ "moe_intermediate_dim": 768,
22
+ "rope_max_wavelength": 10000000,
23
+ "num_key_value_heads": 4,
24
+ "rope_scaling_factor": 1.0,
25
+ "layer_norm_epsilon": 1e-06,
26
+ "dropout": 0,
27
+ "tie_word_embeddings": false,
28
+ "sliding_window_size": null,
29
+ "num_experts": 128,
30
+ "top_k": 8,
31
+ "norm_top_k_prob": true,
32
+ "decoder_sparse_step": 1,
33
+ "mlp_only_layers": [],
34
+ "router_aux_loss_coefficient": 0.001
35
+ },
36
+ "registered_name": "keras_hub>Qwen3MoeBackbone"
37
+ }
metadata.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "keras_version": "3.12.0",
3
+ "keras_hub_version": "0.26.0.dev0",
4
+ "parameter_count": 30532122624,
5
+ "date_saved": "2026-02-01@23:19:19",
6
+ "tasks": [
7
+ "CausalLM"
8
+ ]
9
+ }
model.weights.json ADDED
@@ -0,0 +1,587 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 122128490496.0
4
+ },
5
+ "weight_map": {
6
+ "/layers/reversible_embedding/vars": [
7
+ "model_00000.weights.h5"
8
+ ],
9
+ "/layers/qwen3_moe_transformer_decoder/mlp/expert_bank/vars": [
10
+ "model_00000.weights.h5"
11
+ ],
12
+ "/layers/qwen3_moe_transformer_decoder/mlp/_sparse_feedforward_gate_dense/vars": "model_00000.weights.h5",
13
+ "/layers/qwen3_moe_transformer_decoder/_feedforward_layernorm/vars": "model_00000.weights.h5",
14
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_key_dense/vars": "model_00000.weights.h5",
15
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_key_dense_layer_norm/vars": "model_00000.weights.h5",
16
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_output_dense/vars": "model_00000.weights.h5",
17
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_query_dense/vars": "model_00000.weights.h5",
18
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_query_dense_layer_norm/vars": "model_00000.weights.h5",
19
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layer/_value_dense/vars": "model_00000.weights.h5",
20
+ "/layers/qwen3_moe_transformer_decoder/_self_attention_layernorm/vars": "model_00000.weights.h5",
21
+ "/layers/qwen3_moe_transformer_decoder_1/mlp/expert_bank/vars": [
22
+ "model_00000.weights.h5"
23
+ ],
24
+ "/layers/qwen3_moe_transformer_decoder_1/mlp/_sparse_feedforward_gate_dense/vars": "model_00000.weights.h5",
25
+ "/layers/qwen3_moe_transformer_decoder_1/_feedforward_layernorm/vars": "model_00000.weights.h5",
26
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_key_dense/vars": "model_00000.weights.h5",
27
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_key_dense_layer_norm/vars": "model_00000.weights.h5",
28
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_output_dense/vars": "model_00000.weights.h5",
29
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_query_dense/vars": "model_00000.weights.h5",
30
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_query_dense_layer_norm/vars": "model_00000.weights.h5",
31
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layer/_value_dense/vars": "model_00000.weights.h5",
32
+ "/layers/qwen3_moe_transformer_decoder_1/_self_attention_layernorm/vars": "model_00000.weights.h5",
33
+ "/layers/qwen3_moe_transformer_decoder_2/mlp/expert_bank/vars": [
34
+ "model_00000.weights.h5"
35
+ ],
36
+ "/layers/qwen3_moe_transformer_decoder_2/mlp/_sparse_feedforward_gate_dense/vars": "model_00000.weights.h5",
37
+ "/layers/qwen3_moe_transformer_decoder_2/_feedforward_layernorm/vars": "model_00000.weights.h5",
38
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_key_dense/vars": "model_00000.weights.h5",
39
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_key_dense_layer_norm/vars": "model_00000.weights.h5",
40
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_output_dense/vars": "model_00000.weights.h5",
41
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_query_dense/vars": "model_00000.weights.h5",
42
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_query_dense_layer_norm/vars": "model_00000.weights.h5",
43
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layer/_value_dense/vars": "model_00000.weights.h5",
44
+ "/layers/qwen3_moe_transformer_decoder_2/_self_attention_layernorm/vars": "model_00000.weights.h5",
45
+ "/layers/qwen3_moe_transformer_decoder_3/mlp/expert_bank/vars": [
46
+ "model_00001.weights.h5"
47
+ ],
48
+ "/layers/qwen3_moe_transformer_decoder_3/mlp/_sparse_feedforward_gate_dense/vars": "model_00001.weights.h5",
49
+ "/layers/qwen3_moe_transformer_decoder_3/_feedforward_layernorm/vars": "model_00001.weights.h5",
50
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_key_dense/vars": "model_00001.weights.h5",
51
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_key_dense_layer_norm/vars": "model_00001.weights.h5",
52
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_output_dense/vars": "model_00001.weights.h5",
53
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_query_dense/vars": "model_00001.weights.h5",
54
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_query_dense_layer_norm/vars": "model_00001.weights.h5",
55
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layer/_value_dense/vars": "model_00001.weights.h5",
56
+ "/layers/qwen3_moe_transformer_decoder_3/_self_attention_layernorm/vars": "model_00001.weights.h5",
57
+ "/layers/qwen3_moe_transformer_decoder_4/mlp/expert_bank/vars": [
58
+ "model_00001.weights.h5"
59
+ ],
60
+ "/layers/qwen3_moe_transformer_decoder_4/mlp/_sparse_feedforward_gate_dense/vars": "model_00001.weights.h5",
61
+ "/layers/qwen3_moe_transformer_decoder_4/_feedforward_layernorm/vars": "model_00001.weights.h5",
62
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_key_dense/vars": "model_00001.weights.h5",
63
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_key_dense_layer_norm/vars": "model_00001.weights.h5",
64
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_output_dense/vars": "model_00001.weights.h5",
65
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_query_dense/vars": "model_00001.weights.h5",
66
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_query_dense_layer_norm/vars": "model_00001.weights.h5",
67
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layer/_value_dense/vars": "model_00001.weights.h5",
68
+ "/layers/qwen3_moe_transformer_decoder_4/_self_attention_layernorm/vars": "model_00001.weights.h5",
69
+ "/layers/qwen3_moe_transformer_decoder_5/mlp/expert_bank/vars": [
70
+ "model_00001.weights.h5"
71
+ ],
72
+ "/layers/qwen3_moe_transformer_decoder_5/mlp/_sparse_feedforward_gate_dense/vars": "model_00001.weights.h5",
73
+ "/layers/qwen3_moe_transformer_decoder_5/_feedforward_layernorm/vars": "model_00001.weights.h5",
74
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_key_dense/vars": "model_00001.weights.h5",
75
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_key_dense_layer_norm/vars": "model_00001.weights.h5",
76
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_output_dense/vars": "model_00001.weights.h5",
77
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_query_dense/vars": "model_00001.weights.h5",
78
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_query_dense_layer_norm/vars": "model_00001.weights.h5",
79
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layer/_value_dense/vars": "model_00001.weights.h5",
80
+ "/layers/qwen3_moe_transformer_decoder_5/_self_attention_layernorm/vars": "model_00001.weights.h5",
81
+ "/layers/qwen3_moe_transformer_decoder_6/mlp/expert_bank/vars": [
82
+ "model_00001.weights.h5"
83
+ ],
84
+ "/layers/qwen3_moe_transformer_decoder_6/mlp/_sparse_feedforward_gate_dense/vars": "model_00001.weights.h5",
85
+ "/layers/qwen3_moe_transformer_decoder_6/_feedforward_layernorm/vars": "model_00001.weights.h5",
86
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_key_dense/vars": "model_00001.weights.h5",
87
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_key_dense_layer_norm/vars": "model_00001.weights.h5",
88
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_output_dense/vars": "model_00001.weights.h5",
89
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_query_dense/vars": "model_00001.weights.h5",
90
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_query_dense_layer_norm/vars": "model_00001.weights.h5",
91
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layer/_value_dense/vars": "model_00001.weights.h5",
92
+ "/layers/qwen3_moe_transformer_decoder_6/_self_attention_layernorm/vars": "model_00001.weights.h5",
93
+ "/layers/qwen3_moe_transformer_decoder_7/mlp/expert_bank/vars": [
94
+ "model_00002.weights.h5"
95
+ ],
96
+ "/layers/qwen3_moe_transformer_decoder_7/mlp/_sparse_feedforward_gate_dense/vars": "model_00002.weights.h5",
97
+ "/layers/qwen3_moe_transformer_decoder_7/_feedforward_layernorm/vars": "model_00002.weights.h5",
98
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_key_dense/vars": "model_00002.weights.h5",
99
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_key_dense_layer_norm/vars": "model_00002.weights.h5",
100
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_output_dense/vars": "model_00002.weights.h5",
101
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_query_dense/vars": "model_00002.weights.h5",
102
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_query_dense_layer_norm/vars": "model_00002.weights.h5",
103
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layer/_value_dense/vars": "model_00002.weights.h5",
104
+ "/layers/qwen3_moe_transformer_decoder_7/_self_attention_layernorm/vars": "model_00002.weights.h5",
105
+ "/layers/qwen3_moe_transformer_decoder_8/mlp/expert_bank/vars": [
106
+ "model_00002.weights.h5"
107
+ ],
108
+ "/layers/qwen3_moe_transformer_decoder_8/mlp/_sparse_feedforward_gate_dense/vars": "model_00002.weights.h5",
109
+ "/layers/qwen3_moe_transformer_decoder_8/_feedforward_layernorm/vars": "model_00002.weights.h5",
110
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_key_dense/vars": "model_00002.weights.h5",
111
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_key_dense_layer_norm/vars": "model_00002.weights.h5",
112
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_output_dense/vars": "model_00002.weights.h5",
113
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_query_dense/vars": "model_00002.weights.h5",
114
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_query_dense_layer_norm/vars": "model_00002.weights.h5",
115
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layer/_value_dense/vars": "model_00002.weights.h5",
116
+ "/layers/qwen3_moe_transformer_decoder_8/_self_attention_layernorm/vars": "model_00002.weights.h5",
117
+ "/layers/qwen3_moe_transformer_decoder_9/mlp/expert_bank/vars": [
118
+ "model_00002.weights.h5"
119
+ ],
120
+ "/layers/qwen3_moe_transformer_decoder_9/mlp/_sparse_feedforward_gate_dense/vars": "model_00002.weights.h5",
121
+ "/layers/qwen3_moe_transformer_decoder_9/_feedforward_layernorm/vars": "model_00002.weights.h5",
122
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_key_dense/vars": "model_00002.weights.h5",
123
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_key_dense_layer_norm/vars": "model_00002.weights.h5",
124
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_output_dense/vars": "model_00002.weights.h5",
125
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_query_dense/vars": "model_00002.weights.h5",
126
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_query_dense_layer_norm/vars": "model_00002.weights.h5",
127
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layer/_value_dense/vars": "model_00002.weights.h5",
128
+ "/layers/qwen3_moe_transformer_decoder_9/_self_attention_layernorm/vars": "model_00002.weights.h5",
129
+ "/layers/qwen3_moe_transformer_decoder_10/mlp/expert_bank/vars": [
130
+ "model_00002.weights.h5"
131
+ ],
132
+ "/layers/qwen3_moe_transformer_decoder_10/mlp/_sparse_feedforward_gate_dense/vars": "model_00002.weights.h5",
133
+ "/layers/qwen3_moe_transformer_decoder_10/_feedforward_layernorm/vars": "model_00002.weights.h5",
134
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_key_dense/vars": "model_00002.weights.h5",
135
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_key_dense_layer_norm/vars": "model_00002.weights.h5",
136
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_output_dense/vars": "model_00002.weights.h5",
137
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_query_dense/vars": "model_00002.weights.h5",
138
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_query_dense_layer_norm/vars": "model_00002.weights.h5",
139
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layer/_value_dense/vars": "model_00002.weights.h5",
140
+ "/layers/qwen3_moe_transformer_decoder_10/_self_attention_layernorm/vars": "model_00002.weights.h5",
141
+ "/layers/qwen3_moe_transformer_decoder_11/mlp/expert_bank/vars": [
142
+ "model_00003.weights.h5"
143
+ ],
144
+ "/layers/qwen3_moe_transformer_decoder_11/mlp/_sparse_feedforward_gate_dense/vars": "model_00003.weights.h5",
145
+ "/layers/qwen3_moe_transformer_decoder_11/_feedforward_layernorm/vars": "model_00003.weights.h5",
146
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_key_dense/vars": "model_00003.weights.h5",
147
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_key_dense_layer_norm/vars": "model_00003.weights.h5",
148
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_output_dense/vars": "model_00003.weights.h5",
149
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_query_dense/vars": "model_00003.weights.h5",
150
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_query_dense_layer_norm/vars": "model_00003.weights.h5",
151
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layer/_value_dense/vars": "model_00003.weights.h5",
152
+ "/layers/qwen3_moe_transformer_decoder_11/_self_attention_layernorm/vars": "model_00003.weights.h5",
153
+ "/layers/qwen3_moe_transformer_decoder_12/mlp/expert_bank/vars": [
154
+ "model_00003.weights.h5"
155
+ ],
156
+ "/layers/qwen3_moe_transformer_decoder_12/mlp/_sparse_feedforward_gate_dense/vars": "model_00003.weights.h5",
157
+ "/layers/qwen3_moe_transformer_decoder_12/_feedforward_layernorm/vars": "model_00003.weights.h5",
158
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_key_dense/vars": "model_00003.weights.h5",
159
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_key_dense_layer_norm/vars": "model_00003.weights.h5",
160
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_output_dense/vars": "model_00003.weights.h5",
161
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_query_dense/vars": "model_00003.weights.h5",
162
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_query_dense_layer_norm/vars": "model_00003.weights.h5",
163
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layer/_value_dense/vars": "model_00003.weights.h5",
164
+ "/layers/qwen3_moe_transformer_decoder_12/_self_attention_layernorm/vars": "model_00003.weights.h5",
165
+ "/layers/qwen3_moe_transformer_decoder_13/mlp/expert_bank/vars": [
166
+ "model_00003.weights.h5"
167
+ ],
168
+ "/layers/qwen3_moe_transformer_decoder_13/mlp/_sparse_feedforward_gate_dense/vars": "model_00003.weights.h5",
169
+ "/layers/qwen3_moe_transformer_decoder_13/_feedforward_layernorm/vars": "model_00003.weights.h5",
170
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_key_dense/vars": "model_00003.weights.h5",
171
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_key_dense_layer_norm/vars": "model_00003.weights.h5",
172
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_output_dense/vars": "model_00003.weights.h5",
173
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_query_dense/vars": "model_00003.weights.h5",
174
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_query_dense_layer_norm/vars": "model_00003.weights.h5",
175
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layer/_value_dense/vars": "model_00003.weights.h5",
176
+ "/layers/qwen3_moe_transformer_decoder_13/_self_attention_layernorm/vars": "model_00003.weights.h5",
177
+ "/layers/qwen3_moe_transformer_decoder_14/mlp/expert_bank/vars": [
178
+ "model_00003.weights.h5"
179
+ ],
180
+ "/layers/qwen3_moe_transformer_decoder_14/mlp/_sparse_feedforward_gate_dense/vars": "model_00003.weights.h5",
181
+ "/layers/qwen3_moe_transformer_decoder_14/_feedforward_layernorm/vars": "model_00003.weights.h5",
182
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_key_dense/vars": "model_00003.weights.h5",
183
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_key_dense_layer_norm/vars": "model_00003.weights.h5",
184
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_output_dense/vars": "model_00003.weights.h5",
185
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_query_dense/vars": "model_00003.weights.h5",
186
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_query_dense_layer_norm/vars": "model_00003.weights.h5",
187
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layer/_value_dense/vars": "model_00003.weights.h5",
188
+ "/layers/qwen3_moe_transformer_decoder_14/_self_attention_layernorm/vars": "model_00003.weights.h5",
189
+ "/layers/qwen3_moe_transformer_decoder_15/mlp/expert_bank/vars": [
190
+ "model_00004.weights.h5"
191
+ ],
192
+ "/layers/qwen3_moe_transformer_decoder_15/mlp/_sparse_feedforward_gate_dense/vars": "model_00004.weights.h5",
193
+ "/layers/qwen3_moe_transformer_decoder_15/_feedforward_layernorm/vars": "model_00004.weights.h5",
194
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_key_dense/vars": "model_00004.weights.h5",
195
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_key_dense_layer_norm/vars": "model_00004.weights.h5",
196
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_output_dense/vars": "model_00004.weights.h5",
197
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_query_dense/vars": "model_00004.weights.h5",
198
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_query_dense_layer_norm/vars": "model_00004.weights.h5",
199
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layer/_value_dense/vars": "model_00004.weights.h5",
200
+ "/layers/qwen3_moe_transformer_decoder_15/_self_attention_layernorm/vars": "model_00004.weights.h5",
201
+ "/layers/qwen3_moe_transformer_decoder_16/mlp/expert_bank/vars": [
202
+ "model_00004.weights.h5"
203
+ ],
204
+ "/layers/qwen3_moe_transformer_decoder_16/mlp/_sparse_feedforward_gate_dense/vars": "model_00004.weights.h5",
205
+ "/layers/qwen3_moe_transformer_decoder_16/_feedforward_layernorm/vars": "model_00004.weights.h5",
206
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_key_dense/vars": "model_00004.weights.h5",
207
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_key_dense_layer_norm/vars": "model_00004.weights.h5",
208
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_output_dense/vars": "model_00004.weights.h5",
209
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_query_dense/vars": "model_00004.weights.h5",
210
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_query_dense_layer_norm/vars": "model_00004.weights.h5",
211
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layer/_value_dense/vars": "model_00004.weights.h5",
212
+ "/layers/qwen3_moe_transformer_decoder_16/_self_attention_layernorm/vars": "model_00004.weights.h5",
213
+ "/layers/qwen3_moe_transformer_decoder_17/mlp/expert_bank/vars": [
214
+ "model_00004.weights.h5"
215
+ ],
216
+ "/layers/qwen3_moe_transformer_decoder_17/mlp/_sparse_feedforward_gate_dense/vars": "model_00004.weights.h5",
217
+ "/layers/qwen3_moe_transformer_decoder_17/_feedforward_layernorm/vars": "model_00004.weights.h5",
218
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_key_dense/vars": "model_00004.weights.h5",
219
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_key_dense_layer_norm/vars": "model_00004.weights.h5",
220
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_output_dense/vars": "model_00004.weights.h5",
221
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_query_dense/vars": "model_00004.weights.h5",
222
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_query_dense_layer_norm/vars": "model_00004.weights.h5",
223
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layer/_value_dense/vars": "model_00004.weights.h5",
224
+ "/layers/qwen3_moe_transformer_decoder_17/_self_attention_layernorm/vars": "model_00004.weights.h5",
225
+ "/layers/qwen3_moe_transformer_decoder_18/mlp/expert_bank/vars": [
226
+ "model_00004.weights.h5"
227
+ ],
228
+ "/layers/qwen3_moe_transformer_decoder_18/mlp/_sparse_feedforward_gate_dense/vars": "model_00004.weights.h5",
229
+ "/layers/qwen3_moe_transformer_decoder_18/_feedforward_layernorm/vars": "model_00004.weights.h5",
230
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_key_dense/vars": "model_00004.weights.h5",
231
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_key_dense_layer_norm/vars": "model_00004.weights.h5",
232
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_output_dense/vars": "model_00004.weights.h5",
233
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_query_dense/vars": "model_00004.weights.h5",
234
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_query_dense_layer_norm/vars": "model_00004.weights.h5",
235
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layer/_value_dense/vars": "model_00004.weights.h5",
236
+ "/layers/qwen3_moe_transformer_decoder_18/_self_attention_layernorm/vars": "model_00004.weights.h5",
237
+ "/layers/qwen3_moe_transformer_decoder_19/mlp/expert_bank/vars": [
238
+ "model_00005.weights.h5"
239
+ ],
240
+ "/layers/qwen3_moe_transformer_decoder_19/mlp/_sparse_feedforward_gate_dense/vars": "model_00005.weights.h5",
241
+ "/layers/qwen3_moe_transformer_decoder_19/_feedforward_layernorm/vars": "model_00005.weights.h5",
242
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_key_dense/vars": "model_00005.weights.h5",
243
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_key_dense_layer_norm/vars": "model_00005.weights.h5",
244
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_output_dense/vars": "model_00005.weights.h5",
245
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_query_dense/vars": "model_00005.weights.h5",
246
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_query_dense_layer_norm/vars": "model_00005.weights.h5",
247
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layer/_value_dense/vars": "model_00005.weights.h5",
248
+ "/layers/qwen3_moe_transformer_decoder_19/_self_attention_layernorm/vars": "model_00005.weights.h5",
249
+ "/layers/qwen3_moe_transformer_decoder_20/mlp/expert_bank/vars": [
250
+ "model_00005.weights.h5"
251
+ ],
252
+ "/layers/qwen3_moe_transformer_decoder_20/mlp/_sparse_feedforward_gate_dense/vars": "model_00005.weights.h5",
253
+ "/layers/qwen3_moe_transformer_decoder_20/_feedforward_layernorm/vars": "model_00005.weights.h5",
254
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_key_dense/vars": "model_00005.weights.h5",
255
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_key_dense_layer_norm/vars": "model_00005.weights.h5",
256
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_output_dense/vars": "model_00005.weights.h5",
257
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_query_dense/vars": "model_00005.weights.h5",
258
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_query_dense_layer_norm/vars": "model_00005.weights.h5",
259
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layer/_value_dense/vars": "model_00005.weights.h5",
260
+ "/layers/qwen3_moe_transformer_decoder_20/_self_attention_layernorm/vars": "model_00005.weights.h5",
261
+ "/layers/qwen3_moe_transformer_decoder_21/mlp/expert_bank/vars": [
262
+ "model_00005.weights.h5"
263
+ ],
264
+ "/layers/qwen3_moe_transformer_decoder_21/mlp/_sparse_feedforward_gate_dense/vars": "model_00005.weights.h5",
265
+ "/layers/qwen3_moe_transformer_decoder_21/_feedforward_layernorm/vars": "model_00005.weights.h5",
266
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_key_dense/vars": "model_00005.weights.h5",
267
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_key_dense_layer_norm/vars": "model_00005.weights.h5",
268
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_output_dense/vars": "model_00005.weights.h5",
269
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_query_dense/vars": "model_00005.weights.h5",
270
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_query_dense_layer_norm/vars": "model_00005.weights.h5",
271
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layer/_value_dense/vars": "model_00005.weights.h5",
272
+ "/layers/qwen3_moe_transformer_decoder_21/_self_attention_layernorm/vars": "model_00005.weights.h5",
273
+ "/layers/qwen3_moe_transformer_decoder_22/mlp/expert_bank/vars": [
274
+ "model_00005.weights.h5"
275
+ ],
276
+ "/layers/qwen3_moe_transformer_decoder_22/mlp/_sparse_feedforward_gate_dense/vars": "model_00005.weights.h5",
277
+ "/layers/qwen3_moe_transformer_decoder_22/_feedforward_layernorm/vars": "model_00005.weights.h5",
278
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_key_dense/vars": "model_00005.weights.h5",
279
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_key_dense_layer_norm/vars": "model_00005.weights.h5",
280
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_output_dense/vars": "model_00005.weights.h5",
281
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_query_dense/vars": "model_00005.weights.h5",
282
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_query_dense_layer_norm/vars": "model_00005.weights.h5",
283
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layer/_value_dense/vars": "model_00005.weights.h5",
284
+ "/layers/qwen3_moe_transformer_decoder_22/_self_attention_layernorm/vars": "model_00005.weights.h5",
285
+ "/layers/qwen3_moe_transformer_decoder_23/mlp/expert_bank/vars": [
286
+ "model_00006.weights.h5"
287
+ ],
288
+ "/layers/qwen3_moe_transformer_decoder_23/mlp/_sparse_feedforward_gate_dense/vars": "model_00006.weights.h5",
289
+ "/layers/qwen3_moe_transformer_decoder_23/_feedforward_layernorm/vars": "model_00006.weights.h5",
290
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_key_dense/vars": "model_00006.weights.h5",
291
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_key_dense_layer_norm/vars": "model_00006.weights.h5",
292
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_output_dense/vars": "model_00006.weights.h5",
293
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_query_dense/vars": "model_00006.weights.h5",
294
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_query_dense_layer_norm/vars": "model_00006.weights.h5",
295
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layer/_value_dense/vars": "model_00006.weights.h5",
296
+ "/layers/qwen3_moe_transformer_decoder_23/_self_attention_layernorm/vars": "model_00006.weights.h5",
297
+ "/layers/qwen3_moe_transformer_decoder_24/mlp/expert_bank/vars": [
298
+ "model_00006.weights.h5"
299
+ ],
300
+ "/layers/qwen3_moe_transformer_decoder_24/mlp/_sparse_feedforward_gate_dense/vars": "model_00006.weights.h5",
301
+ "/layers/qwen3_moe_transformer_decoder_24/_feedforward_layernorm/vars": "model_00006.weights.h5",
302
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_key_dense/vars": "model_00006.weights.h5",
303
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_key_dense_layer_norm/vars": "model_00006.weights.h5",
304
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_output_dense/vars": "model_00006.weights.h5",
305
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_query_dense/vars": "model_00006.weights.h5",
306
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_query_dense_layer_norm/vars": "model_00006.weights.h5",
307
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layer/_value_dense/vars": "model_00006.weights.h5",
308
+ "/layers/qwen3_moe_transformer_decoder_24/_self_attention_layernorm/vars": "model_00006.weights.h5",
309
+ "/layers/qwen3_moe_transformer_decoder_25/mlp/expert_bank/vars": [
310
+ "model_00006.weights.h5"
311
+ ],
312
+ "/layers/qwen3_moe_transformer_decoder_25/mlp/_sparse_feedforward_gate_dense/vars": "model_00006.weights.h5",
313
+ "/layers/qwen3_moe_transformer_decoder_25/_feedforward_layernorm/vars": "model_00006.weights.h5",
314
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_key_dense/vars": "model_00006.weights.h5",
315
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_key_dense_layer_norm/vars": "model_00006.weights.h5",
316
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_output_dense/vars": "model_00006.weights.h5",
317
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_query_dense/vars": "model_00006.weights.h5",
318
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_query_dense_layer_norm/vars": "model_00006.weights.h5",
319
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layer/_value_dense/vars": "model_00006.weights.h5",
320
+ "/layers/qwen3_moe_transformer_decoder_25/_self_attention_layernorm/vars": "model_00006.weights.h5",
321
+ "/layers/qwen3_moe_transformer_decoder_26/mlp/expert_bank/vars": [
322
+ "model_00006.weights.h5"
323
+ ],
324
+ "/layers/qwen3_moe_transformer_decoder_26/mlp/_sparse_feedforward_gate_dense/vars": "model_00006.weights.h5",
325
+ "/layers/qwen3_moe_transformer_decoder_26/_feedforward_layernorm/vars": "model_00006.weights.h5",
326
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_key_dense/vars": "model_00006.weights.h5",
327
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_key_dense_layer_norm/vars": "model_00006.weights.h5",
328
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_output_dense/vars": "model_00006.weights.h5",
329
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_query_dense/vars": "model_00006.weights.h5",
330
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_query_dense_layer_norm/vars": "model_00006.weights.h5",
331
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layer/_value_dense/vars": "model_00006.weights.h5",
332
+ "/layers/qwen3_moe_transformer_decoder_26/_self_attention_layernorm/vars": "model_00006.weights.h5",
333
+ "/layers/qwen3_moe_transformer_decoder_27/mlp/expert_bank/vars": [
334
+ "model_00007.weights.h5"
335
+ ],
336
+ "/layers/qwen3_moe_transformer_decoder_27/mlp/_sparse_feedforward_gate_dense/vars": "model_00007.weights.h5",
337
+ "/layers/qwen3_moe_transformer_decoder_27/_feedforward_layernorm/vars": "model_00007.weights.h5",
338
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_key_dense/vars": "model_00007.weights.h5",
339
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_key_dense_layer_norm/vars": "model_00007.weights.h5",
340
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_output_dense/vars": "model_00007.weights.h5",
341
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_query_dense/vars": "model_00007.weights.h5",
342
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_query_dense_layer_norm/vars": "model_00007.weights.h5",
343
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layer/_value_dense/vars": "model_00007.weights.h5",
344
+ "/layers/qwen3_moe_transformer_decoder_27/_self_attention_layernorm/vars": "model_00007.weights.h5",
345
+ "/layers/qwen3_moe_transformer_decoder_28/mlp/expert_bank/vars": [
346
+ "model_00007.weights.h5"
347
+ ],
348
+ "/layers/qwen3_moe_transformer_decoder_28/mlp/_sparse_feedforward_gate_dense/vars": "model_00007.weights.h5",
349
+ "/layers/qwen3_moe_transformer_decoder_28/_feedforward_layernorm/vars": "model_00007.weights.h5",
350
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_key_dense/vars": "model_00007.weights.h5",
351
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_key_dense_layer_norm/vars": "model_00007.weights.h5",
352
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_output_dense/vars": "model_00007.weights.h5",
353
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_query_dense/vars": "model_00007.weights.h5",
354
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_query_dense_layer_norm/vars": "model_00007.weights.h5",
355
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layer/_value_dense/vars": "model_00007.weights.h5",
356
+ "/layers/qwen3_moe_transformer_decoder_28/_self_attention_layernorm/vars": "model_00007.weights.h5",
357
+ "/layers/qwen3_moe_transformer_decoder_29/mlp/expert_bank/vars": [
358
+ "model_00007.weights.h5"
359
+ ],
360
+ "/layers/qwen3_moe_transformer_decoder_29/mlp/_sparse_feedforward_gate_dense/vars": "model_00007.weights.h5",
361
+ "/layers/qwen3_moe_transformer_decoder_29/_feedforward_layernorm/vars": "model_00007.weights.h5",
362
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_key_dense/vars": "model_00007.weights.h5",
363
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_key_dense_layer_norm/vars": "model_00007.weights.h5",
364
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_output_dense/vars": "model_00007.weights.h5",
365
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_query_dense/vars": "model_00007.weights.h5",
366
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_query_dense_layer_norm/vars": "model_00007.weights.h5",
367
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layer/_value_dense/vars": "model_00007.weights.h5",
368
+ "/layers/qwen3_moe_transformer_decoder_29/_self_attention_layernorm/vars": "model_00007.weights.h5",
369
+ "/layers/qwen3_moe_transformer_decoder_30/mlp/expert_bank/vars": [
370
+ "model_00007.weights.h5"
371
+ ],
372
+ "/layers/qwen3_moe_transformer_decoder_30/mlp/_sparse_feedforward_gate_dense/vars": "model_00007.weights.h5",
373
+ "/layers/qwen3_moe_transformer_decoder_30/_feedforward_layernorm/vars": "model_00007.weights.h5",
374
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_key_dense/vars": "model_00007.weights.h5",
375
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_key_dense_layer_norm/vars": "model_00007.weights.h5",
376
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_output_dense/vars": "model_00007.weights.h5",
377
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_query_dense/vars": "model_00007.weights.h5",
378
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_query_dense_layer_norm/vars": "model_00007.weights.h5",
379
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layer/_value_dense/vars": "model_00007.weights.h5",
380
+ "/layers/qwen3_moe_transformer_decoder_30/_self_attention_layernorm/vars": "model_00007.weights.h5",
381
+ "/layers/qwen3_moe_transformer_decoder_31/mlp/expert_bank/vars": [
382
+ "model_00008.weights.h5"
383
+ ],
384
+ "/layers/qwen3_moe_transformer_decoder_31/mlp/_sparse_feedforward_gate_dense/vars": "model_00008.weights.h5",
385
+ "/layers/qwen3_moe_transformer_decoder_31/_feedforward_layernorm/vars": "model_00008.weights.h5",
386
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_key_dense/vars": "model_00008.weights.h5",
387
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_key_dense_layer_norm/vars": "model_00008.weights.h5",
388
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_output_dense/vars": "model_00008.weights.h5",
389
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_query_dense/vars": "model_00008.weights.h5",
390
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_query_dense_layer_norm/vars": "model_00008.weights.h5",
391
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layer/_value_dense/vars": "model_00008.weights.h5",
392
+ "/layers/qwen3_moe_transformer_decoder_31/_self_attention_layernorm/vars": "model_00008.weights.h5",
393
+ "/layers/qwen3_moe_transformer_decoder_32/mlp/expert_bank/vars": [
394
+ "model_00008.weights.h5"
395
+ ],
396
+ "/layers/qwen3_moe_transformer_decoder_32/mlp/_sparse_feedforward_gate_dense/vars": "model_00008.weights.h5",
397
+ "/layers/qwen3_moe_transformer_decoder_32/_feedforward_layernorm/vars": "model_00008.weights.h5",
398
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_key_dense/vars": "model_00008.weights.h5",
399
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_key_dense_layer_norm/vars": "model_00008.weights.h5",
400
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_output_dense/vars": "model_00008.weights.h5",
401
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_query_dense/vars": "model_00008.weights.h5",
402
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_query_dense_layer_norm/vars": "model_00008.weights.h5",
403
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layer/_value_dense/vars": "model_00008.weights.h5",
404
+ "/layers/qwen3_moe_transformer_decoder_32/_self_attention_layernorm/vars": "model_00008.weights.h5",
405
+ "/layers/qwen3_moe_transformer_decoder_33/mlp/expert_bank/vars": [
406
+ "model_00008.weights.h5"
407
+ ],
408
+ "/layers/qwen3_moe_transformer_decoder_33/mlp/_sparse_feedforward_gate_dense/vars": "model_00008.weights.h5",
409
+ "/layers/qwen3_moe_transformer_decoder_33/_feedforward_layernorm/vars": "model_00008.weights.h5",
410
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_key_dense/vars": "model_00008.weights.h5",
411
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_key_dense_layer_norm/vars": "model_00008.weights.h5",
412
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_output_dense/vars": "model_00008.weights.h5",
413
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_query_dense/vars": "model_00008.weights.h5",
414
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_query_dense_layer_norm/vars": "model_00008.weights.h5",
415
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layer/_value_dense/vars": "model_00008.weights.h5",
416
+ "/layers/qwen3_moe_transformer_decoder_33/_self_attention_layernorm/vars": "model_00008.weights.h5",
417
+ "/layers/qwen3_moe_transformer_decoder_34/mlp/expert_bank/vars": [
418
+ "model_00008.weights.h5"
419
+ ],
420
+ "/layers/qwen3_moe_transformer_decoder_34/mlp/_sparse_feedforward_gate_dense/vars": "model_00008.weights.h5",
421
+ "/layers/qwen3_moe_transformer_decoder_34/_feedforward_layernorm/vars": "model_00008.weights.h5",
422
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_key_dense/vars": "model_00008.weights.h5",
423
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_key_dense_layer_norm/vars": "model_00008.weights.h5",
424
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_output_dense/vars": "model_00008.weights.h5",
425
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_query_dense/vars": "model_00008.weights.h5",
426
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_query_dense_layer_norm/vars": "model_00008.weights.h5",
427
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layer/_value_dense/vars": "model_00008.weights.h5",
428
+ "/layers/qwen3_moe_transformer_decoder_34/_self_attention_layernorm/vars": "model_00008.weights.h5",
429
+ "/layers/qwen3_moe_transformer_decoder_35/mlp/expert_bank/vars": [
430
+ "model_00009.weights.h5"
431
+ ],
432
+ "/layers/qwen3_moe_transformer_decoder_35/mlp/_sparse_feedforward_gate_dense/vars": "model_00009.weights.h5",
433
+ "/layers/qwen3_moe_transformer_decoder_35/_feedforward_layernorm/vars": "model_00009.weights.h5",
434
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_key_dense/vars": "model_00009.weights.h5",
435
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_key_dense_layer_norm/vars": "model_00009.weights.h5",
436
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_output_dense/vars": "model_00009.weights.h5",
437
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_query_dense/vars": "model_00009.weights.h5",
438
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_query_dense_layer_norm/vars": "model_00009.weights.h5",
439
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layer/_value_dense/vars": "model_00009.weights.h5",
440
+ "/layers/qwen3_moe_transformer_decoder_35/_self_attention_layernorm/vars": "model_00009.weights.h5",
441
+ "/layers/qwen3_moe_transformer_decoder_36/mlp/expert_bank/vars": [
442
+ "model_00009.weights.h5"
443
+ ],
444
+ "/layers/qwen3_moe_transformer_decoder_36/mlp/_sparse_feedforward_gate_dense/vars": "model_00009.weights.h5",
445
+ "/layers/qwen3_moe_transformer_decoder_36/_feedforward_layernorm/vars": "model_00009.weights.h5",
446
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_key_dense/vars": "model_00009.weights.h5",
447
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_key_dense_layer_norm/vars": "model_00009.weights.h5",
448
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_output_dense/vars": "model_00009.weights.h5",
449
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_query_dense/vars": "model_00009.weights.h5",
450
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_query_dense_layer_norm/vars": "model_00009.weights.h5",
451
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layer/_value_dense/vars": "model_00009.weights.h5",
452
+ "/layers/qwen3_moe_transformer_decoder_36/_self_attention_layernorm/vars": "model_00009.weights.h5",
453
+ "/layers/qwen3_moe_transformer_decoder_37/mlp/expert_bank/vars": [
454
+ "model_00009.weights.h5"
455
+ ],
456
+ "/layers/qwen3_moe_transformer_decoder_37/mlp/_sparse_feedforward_gate_dense/vars": "model_00009.weights.h5",
457
+ "/layers/qwen3_moe_transformer_decoder_37/_feedforward_layernorm/vars": "model_00009.weights.h5",
458
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_key_dense/vars": "model_00009.weights.h5",
459
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_key_dense_layer_norm/vars": "model_00009.weights.h5",
460
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_output_dense/vars": "model_00009.weights.h5",
461
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_query_dense/vars": "model_00009.weights.h5",
462
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_query_dense_layer_norm/vars": "model_00009.weights.h5",
463
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layer/_value_dense/vars": "model_00009.weights.h5",
464
+ "/layers/qwen3_moe_transformer_decoder_37/_self_attention_layernorm/vars": "model_00009.weights.h5",
465
+ "/layers/qwen3_moe_transformer_decoder_38/mlp/expert_bank/vars": [
466
+ "model_00009.weights.h5"
467
+ ],
468
+ "/layers/qwen3_moe_transformer_decoder_38/mlp/_sparse_feedforward_gate_dense/vars": "model_00009.weights.h5",
469
+ "/layers/qwen3_moe_transformer_decoder_38/_feedforward_layernorm/vars": "model_00009.weights.h5",
470
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_key_dense/vars": "model_00009.weights.h5",
471
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_key_dense_layer_norm/vars": "model_00009.weights.h5",
472
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_output_dense/vars": "model_00009.weights.h5",
473
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_query_dense/vars": "model_00009.weights.h5",
474
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_query_dense_layer_norm/vars": "model_00009.weights.h5",
475
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layer/_value_dense/vars": "model_00009.weights.h5",
476
+ "/layers/qwen3_moe_transformer_decoder_38/_self_attention_layernorm/vars": "model_00009.weights.h5",
477
+ "/layers/qwen3_moe_transformer_decoder_39/mlp/expert_bank/vars": [
478
+ "model_00010.weights.h5"
479
+ ],
480
+ "/layers/qwen3_moe_transformer_decoder_39/mlp/_sparse_feedforward_gate_dense/vars": "model_00010.weights.h5",
481
+ "/layers/qwen3_moe_transformer_decoder_39/_feedforward_layernorm/vars": "model_00010.weights.h5",
482
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_key_dense/vars": "model_00010.weights.h5",
483
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_key_dense_layer_norm/vars": "model_00010.weights.h5",
484
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_output_dense/vars": "model_00010.weights.h5",
485
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_query_dense/vars": "model_00010.weights.h5",
486
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_query_dense_layer_norm/vars": "model_00010.weights.h5",
487
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layer/_value_dense/vars": "model_00010.weights.h5",
488
+ "/layers/qwen3_moe_transformer_decoder_39/_self_attention_layernorm/vars": "model_00010.weights.h5",
489
+ "/layers/qwen3_moe_transformer_decoder_40/mlp/expert_bank/vars": [
490
+ "model_00010.weights.h5"
491
+ ],
492
+ "/layers/qwen3_moe_transformer_decoder_40/mlp/_sparse_feedforward_gate_dense/vars": "model_00010.weights.h5",
493
+ "/layers/qwen3_moe_transformer_decoder_40/_feedforward_layernorm/vars": "model_00010.weights.h5",
494
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_key_dense/vars": "model_00010.weights.h5",
495
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_key_dense_layer_norm/vars": "model_00010.weights.h5",
496
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_output_dense/vars": "model_00010.weights.h5",
497
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_query_dense/vars": "model_00010.weights.h5",
498
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_query_dense_layer_norm/vars": "model_00010.weights.h5",
499
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layer/_value_dense/vars": "model_00010.weights.h5",
500
+ "/layers/qwen3_moe_transformer_decoder_40/_self_attention_layernorm/vars": "model_00010.weights.h5",
501
+ "/layers/qwen3_moe_transformer_decoder_41/mlp/expert_bank/vars": [
502
+ "model_00010.weights.h5"
503
+ ],
504
+ "/layers/qwen3_moe_transformer_decoder_41/mlp/_sparse_feedforward_gate_dense/vars": "model_00010.weights.h5",
505
+ "/layers/qwen3_moe_transformer_decoder_41/_feedforward_layernorm/vars": "model_00010.weights.h5",
506
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_key_dense/vars": "model_00010.weights.h5",
507
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_key_dense_layer_norm/vars": "model_00010.weights.h5",
508
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_output_dense/vars": "model_00010.weights.h5",
509
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_query_dense/vars": "model_00010.weights.h5",
510
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_query_dense_layer_norm/vars": "model_00010.weights.h5",
511
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layer/_value_dense/vars": "model_00010.weights.h5",
512
+ "/layers/qwen3_moe_transformer_decoder_41/_self_attention_layernorm/vars": "model_00010.weights.h5",
513
+ "/layers/qwen3_moe_transformer_decoder_42/mlp/expert_bank/vars": [
514
+ "model_00010.weights.h5"
515
+ ],
516
+ "/layers/qwen3_moe_transformer_decoder_42/mlp/_sparse_feedforward_gate_dense/vars": "model_00010.weights.h5",
517
+ "/layers/qwen3_moe_transformer_decoder_42/_feedforward_layernorm/vars": "model_00010.weights.h5",
518
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_key_dense/vars": "model_00010.weights.h5",
519
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_key_dense_layer_norm/vars": "model_00010.weights.h5",
520
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_output_dense/vars": "model_00010.weights.h5",
521
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_query_dense/vars": "model_00010.weights.h5",
522
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_query_dense_layer_norm/vars": "model_00010.weights.h5",
523
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layer/_value_dense/vars": "model_00010.weights.h5",
524
+ "/layers/qwen3_moe_transformer_decoder_42/_self_attention_layernorm/vars": "model_00010.weights.h5",
525
+ "/layers/qwen3_moe_transformer_decoder_43/mlp/expert_bank/vars": [
526
+ "model_00011.weights.h5"
527
+ ],
528
+ "/layers/qwen3_moe_transformer_decoder_43/mlp/_sparse_feedforward_gate_dense/vars": "model_00011.weights.h5",
529
+ "/layers/qwen3_moe_transformer_decoder_43/_feedforward_layernorm/vars": "model_00011.weights.h5",
530
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_key_dense/vars": "model_00011.weights.h5",
531
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_key_dense_layer_norm/vars": "model_00011.weights.h5",
532
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_output_dense/vars": "model_00011.weights.h5",
533
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_query_dense/vars": "model_00011.weights.h5",
534
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_query_dense_layer_norm/vars": "model_00011.weights.h5",
535
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layer/_value_dense/vars": "model_00011.weights.h5",
536
+ "/layers/qwen3_moe_transformer_decoder_43/_self_attention_layernorm/vars": "model_00011.weights.h5",
537
+ "/layers/qwen3_moe_transformer_decoder_44/mlp/expert_bank/vars": [
538
+ "model_00011.weights.h5"
539
+ ],
540
+ "/layers/qwen3_moe_transformer_decoder_44/mlp/_sparse_feedforward_gate_dense/vars": "model_00011.weights.h5",
541
+ "/layers/qwen3_moe_transformer_decoder_44/_feedforward_layernorm/vars": "model_00011.weights.h5",
542
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_key_dense/vars": "model_00011.weights.h5",
543
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_key_dense_layer_norm/vars": "model_00011.weights.h5",
544
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_output_dense/vars": "model_00011.weights.h5",
545
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_query_dense/vars": "model_00011.weights.h5",
546
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_query_dense_layer_norm/vars": "model_00011.weights.h5",
547
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layer/_value_dense/vars": "model_00011.weights.h5",
548
+ "/layers/qwen3_moe_transformer_decoder_44/_self_attention_layernorm/vars": "model_00011.weights.h5",
549
+ "/layers/qwen3_moe_transformer_decoder_45/mlp/expert_bank/vars": [
550
+ "model_00011.weights.h5"
551
+ ],
552
+ "/layers/qwen3_moe_transformer_decoder_45/mlp/_sparse_feedforward_gate_dense/vars": "model_00011.weights.h5",
553
+ "/layers/qwen3_moe_transformer_decoder_45/_feedforward_layernorm/vars": "model_00011.weights.h5",
554
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_key_dense/vars": "model_00011.weights.h5",
555
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_key_dense_layer_norm/vars": "model_00011.weights.h5",
556
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_output_dense/vars": "model_00011.weights.h5",
557
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_query_dense/vars": "model_00011.weights.h5",
558
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_query_dense_layer_norm/vars": "model_00011.weights.h5",
559
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layer/_value_dense/vars": "model_00011.weights.h5",
560
+ "/layers/qwen3_moe_transformer_decoder_45/_self_attention_layernorm/vars": "model_00011.weights.h5",
561
+ "/layers/qwen3_moe_transformer_decoder_46/mlp/expert_bank/vars": [
562
+ "model_00011.weights.h5"
563
+ ],
564
+ "/layers/qwen3_moe_transformer_decoder_46/mlp/_sparse_feedforward_gate_dense/vars": "model_00011.weights.h5",
565
+ "/layers/qwen3_moe_transformer_decoder_46/_feedforward_layernorm/vars": "model_00011.weights.h5",
566
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_key_dense/vars": "model_00011.weights.h5",
567
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_key_dense_layer_norm/vars": "model_00011.weights.h5",
568
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_output_dense/vars": "model_00011.weights.h5",
569
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_query_dense/vars": "model_00011.weights.h5",
570
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_query_dense_layer_norm/vars": "model_00011.weights.h5",
571
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layer/_value_dense/vars": "model_00011.weights.h5",
572
+ "/layers/qwen3_moe_transformer_decoder_46/_self_attention_layernorm/vars": "model_00011.weights.h5",
573
+ "/layers/qwen3_moe_transformer_decoder_47/mlp/expert_bank/vars": [
574
+ "model_00012.weights.h5"
575
+ ],
576
+ "/layers/qwen3_moe_transformer_decoder_47/mlp/_sparse_feedforward_gate_dense/vars": "model_00012.weights.h5",
577
+ "/layers/qwen3_moe_transformer_decoder_47/_feedforward_layernorm/vars": "model_00012.weights.h5",
578
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_key_dense/vars": "model_00012.weights.h5",
579
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_key_dense_layer_norm/vars": "model_00012.weights.h5",
580
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_output_dense/vars": "model_00012.weights.h5",
581
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_query_dense/vars": "model_00012.weights.h5",
582
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_query_dense_layer_norm/vars": "model_00012.weights.h5",
583
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layer/_value_dense/vars": "model_00012.weights.h5",
584
+ "/layers/qwen3_moe_transformer_decoder_47/_self_attention_layernorm/vars": "model_00012.weights.h5",
585
+ "/layers/qwen3_moe_layer_norm/vars": "model_00012.weights.h5"
586
+ }
587
+ }
model_00000.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22d8c83a8c1a7ffde37958b0c3a291be2356e969861bdff85ad4799af6167117
3
+ size 9966901888
model_00001.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4de3c2fe4b4540601aa2cd18dae15238742c02f08dade13fbb16dfa2a9026c00
3
+ size 9970093320
model_00002.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c2905bc865be586fb5541b979017e32f13c13197faddd70ad0ec9c10b63944d8
3
+ size 9970093320
model_00003.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d4d39732c35be9e828cc5ed0266f9657081a5c33b1fab874148302737dce118
3
+ size 9970093320
model_00004.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:338193664dc80341ccea52ffc7865b397f18a9d6f5f81b159bbd91f5327c9359
3
+ size 9970093320
model_00005.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d0db94cf7926fea7bdef1d8ccf09abebd24ab465acd332e3a0727576ba6dd12
3
+ size 9970093320
model_00006.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8c0bf9be0a71aa98212425faaef2ce13d6bf032fb5ab5f06a942218aabd0c176
3
+ size 9970093320
model_00007.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bdf94e228502a73e42d9856ac2050b8300b36e8ca0a3aae04610f584bd869a74
3
+ size 9970093320
model_00008.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:74848db9ffdf403e048859f9bae245ad48203b61016c7fa66bea44f722437d6c
3
+ size 9970093320
model_00009.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8b65a1439644af108d839fcac8b7b09325aebecdd8a32fd7debece75a9d3925d
3
+ size 9970093320
model_00010.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e9cee03d751b97d5bdfde7318c69f3b9aa59fc963e5ab1c5aeccb0e740ac3249
3
+ size 9970093320
model_00011.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d8c81c865bcb9b176f67400f5bafcae87613517876954643654a540a7b6c9b3
3
+ size 9970093320
model_00012.weights.h5 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d82f93b08ee353c053c95b11639834db1a72d299da949e68e6df4e2c4ef0089
3
+ size 2492536952
preprocessor.json ADDED
@@ -0,0 +1,69 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm_preprocessor",
3
+ "class_name": "Qwen3MoeCausalLMPreprocessor",
4
+ "config": {
5
+ "name": "qwen3_moe_causal_lm_preprocessor_2",
6
+ "trainable": true,
7
+ "dtype": {
8
+ "module": "keras",
9
+ "class_name": "DTypePolicy",
10
+ "config": {
11
+ "name": "float32"
12
+ },
13
+ "registered_name": null
14
+ },
15
+ "tokenizer": {
16
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer",
17
+ "class_name": "Qwen3MoeTokenizer",
18
+ "config": {
19
+ "name": "qwen3_moe_tokenizer",
20
+ "trainable": true,
21
+ "dtype": {
22
+ "module": "keras",
23
+ "class_name": "DTypePolicy",
24
+ "config": {
25
+ "name": "int32"
26
+ },
27
+ "registered_name": null
28
+ },
29
+ "config_file": "tokenizer.json",
30
+ "sequence_length": null,
31
+ "add_prefix_space": false,
32
+ "unsplittable_tokens": [
33
+ "<tool_response>",
34
+ "</think>",
35
+ "<|box_end|>",
36
+ "</tool_response>",
37
+ "<|fim_suffix|>",
38
+ "<|quad_start|>",
39
+ "<tool_call>",
40
+ "<think>",
41
+ "<|vision_pad|>",
42
+ "<|file_sep|>",
43
+ "<|box_start|>",
44
+ "<|im_start|>",
45
+ "<|endoftext|>",
46
+ "</tool_call>",
47
+ "<|object_ref_start|>",
48
+ "<|repo_name|>",
49
+ "<|vision_start|>",
50
+ "<|vision_end|>",
51
+ "<|object_ref_end|>",
52
+ "<|quad_end|>",
53
+ "<|fim_middle|>",
54
+ "<|video_pad|>",
55
+ "<|im_end|>",
56
+ "<|image_pad|>",
57
+ "<|fim_prefix|>",
58
+ "<|fim_pad|>"
59
+ ]
60
+ },
61
+ "registered_name": "keras_hub>Qwen3MoeTokenizer"
62
+ },
63
+ "config_file": "preprocessor.json",
64
+ "sequence_length": 1024,
65
+ "add_start_token": true,
66
+ "add_end_token": true
67
+ },
68
+ "registered_name": "keras_hub>Qwen3MoeCausalLMPreprocessor"
69
+ }
task.json ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm",
3
+ "class_name": "Qwen3MoeCausalLM",
4
+ "config": {
5
+ "backbone": {
6
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_backbone",
7
+ "class_name": "Qwen3MoeBackbone",
8
+ "config": {
9
+ "name": "qwen3_moe_backbone",
10
+ "trainable": true,
11
+ "dtype": {
12
+ "module": "keras",
13
+ "class_name": "DTypePolicy",
14
+ "config": {
15
+ "name": "float32"
16
+ },
17
+ "registered_name": null
18
+ },
19
+ "vocabulary_size": 151936,
20
+ "num_layers": 48,
21
+ "num_query_heads": 32,
22
+ "head_dim": 128,
23
+ "hidden_dim": 2048,
24
+ "intermediate_dim": 6144,
25
+ "moe_intermediate_dim": 768,
26
+ "rope_max_wavelength": 10000000,
27
+ "num_key_value_heads": 4,
28
+ "rope_scaling_factor": 1.0,
29
+ "layer_norm_epsilon": 1e-06,
30
+ "dropout": 0,
31
+ "tie_word_embeddings": false,
32
+ "sliding_window_size": null,
33
+ "num_experts": 128,
34
+ "top_k": 8,
35
+ "norm_top_k_prob": true,
36
+ "decoder_sparse_step": 1,
37
+ "mlp_only_layers": [],
38
+ "router_aux_loss_coefficient": 0.001
39
+ },
40
+ "registered_name": "keras_hub>Qwen3MoeBackbone"
41
+ },
42
+ "preprocessor": {
43
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm_preprocessor",
44
+ "class_name": "Qwen3MoeCausalLMPreprocessor",
45
+ "config": {
46
+ "name": "qwen3_moe_causal_lm_preprocessor_2",
47
+ "trainable": true,
48
+ "dtype": {
49
+ "module": "keras",
50
+ "class_name": "DTypePolicy",
51
+ "config": {
52
+ "name": "float32"
53
+ },
54
+ "registered_name": null
55
+ },
56
+ "tokenizer": {
57
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer",
58
+ "class_name": "Qwen3MoeTokenizer",
59
+ "config": {
60
+ "name": "qwen3_moe_tokenizer",
61
+ "trainable": true,
62
+ "dtype": {
63
+ "module": "keras",
64
+ "class_name": "DTypePolicy",
65
+ "config": {
66
+ "name": "int32"
67
+ },
68
+ "registered_name": null
69
+ },
70
+ "config_file": "tokenizer.json",
71
+ "sequence_length": null,
72
+ "add_prefix_space": false,
73
+ "unsplittable_tokens": [
74
+ "<tool_response>",
75
+ "</think>",
76
+ "<|box_end|>",
77
+ "</tool_response>",
78
+ "<|fim_suffix|>",
79
+ "<|quad_start|>",
80
+ "<tool_call>",
81
+ "<think>",
82
+ "<|vision_pad|>",
83
+ "<|file_sep|>",
84
+ "<|box_start|>",
85
+ "<|im_start|>",
86
+ "<|endoftext|>",
87
+ "</tool_call>",
88
+ "<|object_ref_start|>",
89
+ "<|repo_name|>",
90
+ "<|vision_start|>",
91
+ "<|vision_end|>",
92
+ "<|object_ref_end|>",
93
+ "<|quad_end|>",
94
+ "<|fim_middle|>",
95
+ "<|video_pad|>",
96
+ "<|im_end|>",
97
+ "<|image_pad|>",
98
+ "<|fim_prefix|>",
99
+ "<|fim_pad|>"
100
+ ]
101
+ },
102
+ "registered_name": "keras_hub>Qwen3MoeTokenizer"
103
+ },
104
+ "config_file": "preprocessor.json",
105
+ "sequence_length": 1024,
106
+ "add_start_token": true,
107
+ "add_end_token": true
108
+ },
109
+ "registered_name": "keras_hub>Qwen3MoeCausalLMPreprocessor"
110
+ },
111
+ "name": "qwen3_moe_causal_lm"
112
+ },
113
+ "registered_name": "keras_hub>Qwen3MoeCausalLM"
114
+ }
tokenizer.json ADDED
@@ -0,0 +1,48 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "module": "keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer",
3
+ "class_name": "Qwen3MoeTokenizer",
4
+ "config": {
5
+ "name": "qwen3_moe_tokenizer",
6
+ "trainable": true,
7
+ "dtype": {
8
+ "module": "keras",
9
+ "class_name": "DTypePolicy",
10
+ "config": {
11
+ "name": "int32"
12
+ },
13
+ "registered_name": null
14
+ },
15
+ "config_file": "tokenizer.json",
16
+ "sequence_length": null,
17
+ "add_prefix_space": false,
18
+ "unsplittable_tokens": [
19
+ "<tool_response>",
20
+ "</think>",
21
+ "<|box_end|>",
22
+ "</tool_response>",
23
+ "<|fim_suffix|>",
24
+ "<|quad_start|>",
25
+ "<tool_call>",
26
+ "<think>",
27
+ "<|vision_pad|>",
28
+ "<|file_sep|>",
29
+ "<|box_start|>",
30
+ "<|im_start|>",
31
+ "<|endoftext|>",
32
+ "</tool_call>",
33
+ "<|object_ref_start|>",
34
+ "<|repo_name|>",
35
+ "<|vision_start|>",
36
+ "<|vision_end|>",
37
+ "<|object_ref_end|>",
38
+ "<|quad_end|>",
39
+ "<|fim_middle|>",
40
+ "<|video_pad|>",
41
+ "<|im_end|>",
42
+ "<|image_pad|>",
43
+ "<|fim_prefix|>",
44
+ "<|fim_pad|>"
45
+ ]
46
+ },
47
+ "registered_name": "keras_hub>Qwen3MoeTokenizer"
48
+ }