Hugging Face's logo

avlp12
/
ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic

Image-Text-to-Text
MLX
Safetensors
English
Korean
Chinese
qwen3_5
quantized
mixed-precision
apple-silicon
token-efficient
efficient-thinking
conversational
Eval Results
4-bit precision
Model card Files
xet
 Community
avlp12 commited on
Commit
3b76368
·
verified ·
1 Parent(s): 71a6a45

Upload folder using huggingface_hub

Browse files
Files changed (20) hide show
  1. .eval_results/gpqa.yaml +7 -0
  2. .eval_results/gsm8k.yaml +7 -0
  3. .eval_results/mmlu-pro.yaml +7 -0
  4. README.md +11 -10
  5. assets/fig_gpqa_curve.png +0 -0
  6. assets/fig_showdown.png +0 -0
  7. config.json +707 -1
  8. configuration.json +1 -0
  9. merge_verify_report.json +2225 -0
  10. model-00003-of-00003.safetensors → model-00001-of-00004.safetensors +2 -2
  11. model-00001-of-00003.safetensors → model-00002-of-00004.safetensors +2 -2
  12. model-00002-of-00003.safetensors → model-00003-of-00004.safetensors +2 -2
  13. model-00004-of-00004.safetensors +3 -0
  14. model.safetensors.index.json +0 -0
  15. preprocessor_config.json +21 -0
  16. processor_config.json +50 -0
  17. tokenizer_config.json +1 -1
  18. video_preprocessor_config.json +21 -0
  19. vocab.json +0 -0
  20. weight_check_report.json +43 -0
.eval_results/gpqa.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ - dataset:
2
+ id: Idavidrein/gpqa
3
+ task_id: diamond
4
+ value: 83.8
5
+ source:
6
+ url: https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B
7
+ name: Model Card
.eval_results/gsm8k.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ - dataset:
2
+ id: openai/gsm8k
3
+ task_id: gsm8k
4
+ value: 96.5
5
+ source:
6
+ url: https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B
7
+ name: Model Card
.eval_results/mmlu-pro.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ - dataset:
2
+ id: TIGER-Lab/MMLU-Pro
3
+ task_id: mmlu_pro
4
+ value: 85.4
5
+ source:
6
+ url: https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B
7
+ name: Model Card
README.md CHANGED
@@ -7,7 +7,7 @@ license: apache-2.0
7
  base_model: bottlecapai/ThinkingCap-Qwen3.6-27B
8
  base_model_relation: quantized
9
  library_name: mlx
10
- pipeline_tag: text-generation
11
  tags:
12
  - mlx
13
  - qwen3_5
@@ -25,8 +25,8 @@ Apple Silicon (MLX) **mixed-precision** quantization of [bottlecapai/ThinkingCap
25
 
26
  | build | weights | GPQA-D* | avg think tokens | decode† | peak mem |
27
  |---|---|---|---|---|---|
28
- | **`main` (this)** | **15.47 GB** | **83.3%** | **2,838** | 36.5 tok/s | 15.8 GB |
29
- | [`vision`](https://huggingface.co/avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic/tree/vision) — multimodal | 16.13 GB | (same text weights) | — | 35.6 tok/s | 16.7 GB |
30
  | [BottleCapAI Q4_K_M GGUF (official)](https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B-GGUF) | 16.81 GB | 82.3% | 3,654 | — | — |
31
  | BottleCapAI bf16 (original) | 55.56 GB | 83.8% | 3,019 | — | — |
32
 
@@ -53,17 +53,18 @@ reasoning chains accumulate quantization drift token by token; ThinkingCap's
53
 
54
  ![GPQA vs bits-per-weight: vanilla quants degrade steeply; the ThinkingCap quant sits on the bf16 line](https://huggingface.co/avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic/resolve/main/assets/fig_gpqa_curve.png)
55
 
56
- ## Multimodal (`vision` branch)
57
 
58
- ThinkingCap ships with Qwen3.6's full vision tower; `main` (mlx-lm) is text-only, but
59
- the `vision` branch restores it: **identical 4.6bpw text weights** (so the GPQA number
60
- carries over) + the vision tower at 8-bit. 16.13 GB total — still smaller than the
61
- 16.81 GB text-only GGUF. Verified with OCR + visual-reasoning tests (EN/KO).
 
62
 
63
  ```bash
64
  pip install mlx-vlm
65
  python -m mlx_vlm generate \
66
- --model avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic --revision vision \
67
  --image photo.jpg --prompt "이미지의 텍스트를 읽어줘." --max-tokens 300
68
  ```
69
 
@@ -105,7 +106,7 @@ teacher-close); conversion-time weights ship as-is.
105
 
106
  ## Notes & provenance
107
 
108
- - `main` is text-only (mlx-lm path); the `vision` branch carries the full multimodal stack (mlx-vlm).
109
  - Eval artifacts: same 198-question GPQA protocol for every row of the table above.
110
  - Built on an M3 Ultra 512 GB · mlx-lm 0.31.3 · mlx 0.31.2 · 2026-07.
111
  - Siblings: [Qwen3.6-27B-Alis-MLX-Dynamic](https://huggingface.co/avlp12/Qwen3.6-27B-Alis-MLX-Dynamic) (vanilla golden + 5.5bpw reasoning tier), [Qwen3.6-35B-A3B-Alis-MLX-Dynamic](https://huggingface.co/avlp12/Qwen3.6-35B-A3B-Alis-MLX-Dynamic) (MoE, 102 tok/s).
 
7
  base_model: bottlecapai/ThinkingCap-Qwen3.6-27B
8
  base_model_relation: quantized
9
  library_name: mlx
10
+ pipeline_tag: image-text-to-text
11
  tags:
12
  - mlx
13
  - qwen3_5
 
25
 
26
  | build | weights | GPQA-D* | avg think tokens | decode† | peak mem |
27
  |---|---|---|---|---|---|
28
+ | **`main` (this)** — multimodal | **16.13 GB** | **83.3%**¹ | **2,838** | 35.6 tok/s | 16.7 GB |
29
+ | [`text`](https://huggingface.co/avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic/tree/text) — text-only | 15.47 GB | 83.3% | 2,838 | 36.5 tok/s | 15.8 GB |
30
  | [BottleCapAI Q4_K_M GGUF (official)](https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B-GGUF) | 16.81 GB | 82.3% | 3,654 | — | — |
31
  | BottleCapAI bf16 (original) | 55.56 GB | 83.8% | 3,019 | — | — |
32
 
 
53
 
54
  ![GPQA vs bits-per-weight: vanilla quants degrade steeply; the ThinkingCap quant sits on the bf16 line](https://huggingface.co/avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic/resolve/main/assets/fig_gpqa_curve.png)
55
 
56
+ ## Multimodal (default)
57
 
58
+ `main` carries the full multimodal stack: the 4.6bpw text weights (¹ GPQA measured on the
59
+ identical text weights) + the vision tower at 8-bit. 16.13 GB total — still smaller than
60
+ the 16.81 GB *text-only* GGUF. Verified with OCR + visual-reasoning tests (EN/KO).
61
+ Text-only users: `main` loads fine under plain mlx-lm (vision weights are skipped), or
62
+ use the `text` branch to save 0.66 GB.
63
 
64
  ```bash
65
  pip install mlx-vlm
66
  python -m mlx_vlm generate \
67
+ --model avlp12/ThinkingCap-Qwen3.6-27B-Alis-MLX-Dynamic \
68
  --image photo.jpg --prompt "이미지의 텍스트를 읽어줘." --max-tokens 300
69
  ```
70
 
 
106
 
107
  ## Notes & provenance
108
 
109
+ - `main` carries the full multimodal stack (mlx-vlm); the `text` branch is the mlx-lm text-only artifact.
110
  - Eval artifacts: same 198-question GPQA protocol for every row of the table above.
111
  - Built on an M3 Ultra 512 GB · mlx-lm 0.31.3 · mlx 0.31.2 · 2026-07.
112
  - Siblings: [Qwen3.6-27B-Alis-MLX-Dynamic](https://huggingface.co/avlp12/Qwen3.6-27B-Alis-MLX-Dynamic) (vanilla golden + 5.5bpw reasoning tier), [Qwen3.6-35B-A3B-Alis-MLX-Dynamic](https://huggingface.co/avlp12/Qwen3.6-35B-A3B-Alis-MLX-Dynamic) (MoE, 102 tok/s).
assets/fig_gpqa_curve.png DELETED
Binary file (77.6 kB)
 
assets/fig_showdown.png DELETED
Binary file (77.5 kB)
 
config.json CHANGED
@@ -2,11 +2,25 @@
2
  "architectures": [
3
  "Qwen3_5ForConditionalGeneration"
4
  ],
 
5
  "dtype": "bfloat16",
6
  "eos_token_id": [
7
  248046,
8
  248044
9
  ],
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  "image_token_id": 248056,
11
  "language_model_only": false,
12
  "model_type": "qwen3_5",
@@ -14,6 +28,342 @@
14
  "group_size": 64,
15
  "bits": 4,
16
  "mode": "affine",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
  "language_model.model.embed_tokens": {
18
  "bits": 4,
19
  "group_size": 64,
@@ -2509,6 +2859,342 @@
2509
  "group_size": 64,
2510
  "bits": 4,
2511
  "mode": "affine",
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2512
  "language_model.model.embed_tokens": {
2513
  "bits": 4,
2514
  "group_size": 64,
@@ -5000,6 +5686,7 @@
5000
  "mode": "affine"
5001
  }
5002
  },
 
5003
  "text_config": {
5004
  "attention_bias": false,
5005
  "attention_dropout": 0.0,
@@ -5105,15 +5792,34 @@
5105
  ],
5106
  "partial_rotary_factor": 0.25,
5107
  "rope_theta": 10000000,
5108
- "type": "default"
5109
  },
5110
  "tie_word_embeddings": false,
5111
  "use_cache": true,
5112
  "vocab_size": 248320
5113
  },
5114
  "tie_word_embeddings": false,
 
 
5115
  "transformers_version": "5.7.0",
5116
  "video_token_id": 248057,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5117
  "vision_end_token_id": 248054,
5118
  "vision_start_token_id": 248053
5119
  }
 
2
  "architectures": [
3
  "Qwen3_5ForConditionalGeneration"
4
  ],
5
+ "do_sample": true,
6
  "dtype": "bfloat16",
7
  "eos_token_id": [
8
  248046,
9
  248044
10
  ],
11
+ "generation_config": {
12
+ "bos_token_id": 248044,
13
+ "do_sample": true,
14
+ "eos_token_id": [
15
+ 248046,
16
+ 248044
17
+ ],
18
+ "pad_token_id": 248044,
19
+ "temperature": 1.0,
20
+ "top_k": 20,
21
+ "top_p": 0.95,
22
+ "transformers_version": "5.7.0"
23
+ },
24
  "image_token_id": 248056,
25
  "language_model_only": false,
26
  "model_type": "qwen3_5",
 
28
  "group_size": 64,
29
  "bits": 4,
30
  "mode": "affine",
31
+ "vision_tower.pos_embed": {
32
+ "group_size": 64,
33
+ "bits": 8
34
+ },
35
+ "vision_tower.blocks.0.attn.qkv": {
36
+ "group_size": 64,
37
+ "bits": 8
38
+ },
39
+ "vision_tower.blocks.0.attn.proj": {
40
+ "group_size": 64,
41
+ "bits": 8
42
+ },
43
+ "vision_tower.blocks.0.mlp.linear_fc1": {
44
+ "group_size": 64,
45
+ "bits": 8
46
+ },
47
+ "vision_tower.blocks.1.attn.qkv": {
48
+ "group_size": 64,
49
+ "bits": 8
50
+ },
51
+ "vision_tower.blocks.1.attn.proj": {
52
+ "group_size": 64,
53
+ "bits": 8
54
+ },
55
+ "vision_tower.blocks.1.mlp.linear_fc1": {
56
+ "group_size": 64,
57
+ "bits": 8
58
+ },
59
+ "vision_tower.blocks.2.attn.qkv": {
60
+ "group_size": 64,
61
+ "bits": 8
62
+ },
63
+ "vision_tower.blocks.2.attn.proj": {
64
+ "group_size": 64,
65
+ "bits": 8
66
+ },
67
+ "vision_tower.blocks.2.mlp.linear_fc1": {
68
+ "group_size": 64,
69
+ "bits": 8
70
+ },
71
+ "vision_tower.blocks.3.attn.qkv": {
72
+ "group_size": 64,
73
+ "bits": 8
74
+ },
75
+ "vision_tower.blocks.3.attn.proj": {
76
+ "group_size": 64,
77
+ "bits": 8
78
+ },
79
+ "vision_tower.blocks.3.mlp.linear_fc1": {
80
+ "group_size": 64,
81
+ "bits": 8
82
+ },
83
+ "vision_tower.blocks.4.attn.qkv": {
84
+ "group_size": 64,
85
+ "bits": 8
86
+ },
87
+ "vision_tower.blocks.4.attn.proj": {
88
+ "group_size": 64,
89
+ "bits": 8
90
+ },
91
+ "vision_tower.blocks.4.mlp.linear_fc1": {
92
+ "group_size": 64,
93
+ "bits": 8
94
+ },
95
+ "vision_tower.blocks.5.attn.qkv": {
96
+ "group_size": 64,
97
+ "bits": 8
98
+ },
99
+ "vision_tower.blocks.5.attn.proj": {
100
+ "group_size": 64,
101
+ "bits": 8
102
+ },
103
+ "vision_tower.blocks.5.mlp.linear_fc1": {
104
+ "group_size": 64,
105
+ "bits": 8
106
+ },
107
+ "vision_tower.blocks.6.attn.qkv": {
108
+ "group_size": 64,
109
+ "bits": 8
110
+ },
111
+ "vision_tower.blocks.6.attn.proj": {
112
+ "group_size": 64,
113
+ "bits": 8
114
+ },
115
+ "vision_tower.blocks.6.mlp.linear_fc1": {
116
+ "group_size": 64,
117
+ "bits": 8
118
+ },
119
+ "vision_tower.blocks.7.attn.qkv": {
120
+ "group_size": 64,
121
+ "bits": 8
122
+ },
123
+ "vision_tower.blocks.7.attn.proj": {
124
+ "group_size": 64,
125
+ "bits": 8
126
+ },
127
+ "vision_tower.blocks.7.mlp.linear_fc1": {
128
+ "group_size": 64,
129
+ "bits": 8
130
+ },
131
+ "vision_tower.blocks.8.attn.qkv": {
132
+ "group_size": 64,
133
+ "bits": 8
134
+ },
135
+ "vision_tower.blocks.8.attn.proj": {
136
+ "group_size": 64,
137
+ "bits": 8
138
+ },
139
+ "vision_tower.blocks.8.mlp.linear_fc1": {
140
+ "group_size": 64,
141
+ "bits": 8
142
+ },
143
+ "vision_tower.blocks.9.attn.qkv": {
144
+ "group_size": 64,
145
+ "bits": 8
146
+ },
147
+ "vision_tower.blocks.9.attn.proj": {
148
+ "group_size": 64,
149
+ "bits": 8
150
+ },
151
+ "vision_tower.blocks.9.mlp.linear_fc1": {
152
+ "group_size": 64,
153
+ "bits": 8
154
+ },
155
+ "vision_tower.blocks.10.attn.qkv": {
156
+ "group_size": 64,
157
+ "bits": 8
158
+ },
159
+ "vision_tower.blocks.10.attn.proj": {
160
+ "group_size": 64,
161
+ "bits": 8
162
+ },
163
+ "vision_tower.blocks.10.mlp.linear_fc1": {
164
+ "group_size": 64,
165
+ "bits": 8
166
+ },
167
+ "vision_tower.blocks.11.attn.qkv": {
168
+ "group_size": 64,
169
+ "bits": 8
170
+ },
171
+ "vision_tower.blocks.11.attn.proj": {
172
+ "group_size": 64,
173
+ "bits": 8
174
+ },
175
+ "vision_tower.blocks.11.mlp.linear_fc1": {
176
+ "group_size": 64,
177
+ "bits": 8
178
+ },
179
+ "vision_tower.blocks.12.attn.qkv": {
180
+ "group_size": 64,
181
+ "bits": 8
182
+ },
183
+ "vision_tower.blocks.12.attn.proj": {
184
+ "group_size": 64,
185
+ "bits": 8
186
+ },
187
+ "vision_tower.blocks.12.mlp.linear_fc1": {
188
+ "group_size": 64,
189
+ "bits": 8
190
+ },
191
+ "vision_tower.blocks.13.attn.qkv": {
192
+ "group_size": 64,
193
+ "bits": 8
194
+ },
195
+ "vision_tower.blocks.13.attn.proj": {
196
+ "group_size": 64,
197
+ "bits": 8
198
+ },
199
+ "vision_tower.blocks.13.mlp.linear_fc1": {
200
+ "group_size": 64,
201
+ "bits": 8
202
+ },
203
+ "vision_tower.blocks.14.attn.qkv": {
204
+ "group_size": 64,
205
+ "bits": 8
206
+ },
207
+ "vision_tower.blocks.14.attn.proj": {
208
+ "group_size": 64,
209
+ "bits": 8
210
+ },
211
+ "vision_tower.blocks.14.mlp.linear_fc1": {
212
+ "group_size": 64,
213
+ "bits": 8
214
+ },
215
+ "vision_tower.blocks.15.attn.qkv": {
216
+ "group_size": 64,
217
+ "bits": 8
218
+ },
219
+ "vision_tower.blocks.15.attn.proj": {
220
+ "group_size": 64,
221
+ "bits": 8
222
+ },
223
+ "vision_tower.blocks.15.mlp.linear_fc1": {
224
+ "group_size": 64,
225
+ "bits": 8
226
+ },
227
+ "vision_tower.blocks.16.attn.qkv": {
228
+ "group_size": 64,
229
+ "bits": 8
230
+ },
231
+ "vision_tower.blocks.16.attn.proj": {
232
+ "group_size": 64,
233
+ "bits": 8
234
+ },
235
+ "vision_tower.blocks.16.mlp.linear_fc1": {
236
+ "group_size": 64,
237
+ "bits": 8
238
+ },
239
+ "vision_tower.blocks.17.attn.qkv": {
240
+ "group_size": 64,
241
+ "bits": 8
242
+ },
243
+ "vision_tower.blocks.17.attn.proj": {
244
+ "group_size": 64,
245
+ "bits": 8
246
+ },
247
+ "vision_tower.blocks.17.mlp.linear_fc1": {
248
+ "group_size": 64,
249
+ "bits": 8
250
+ },
251
+ "vision_tower.blocks.18.attn.qkv": {
252
+ "group_size": 64,
253
+ "bits": 8
254
+ },
255
+ "vision_tower.blocks.18.attn.proj": {
256
+ "group_size": 64,
257
+ "bits": 8
258
+ },
259
+ "vision_tower.blocks.18.mlp.linear_fc1": {
260
+ "group_size": 64,
261
+ "bits": 8
262
+ },
263
+ "vision_tower.blocks.19.attn.qkv": {
264
+ "group_size": 64,
265
+ "bits": 8
266
+ },
267
+ "vision_tower.blocks.19.attn.proj": {
268
+ "group_size": 64,
269
+ "bits": 8
270
+ },
271
+ "vision_tower.blocks.19.mlp.linear_fc1": {
272
+ "group_size": 64,
273
+ "bits": 8
274
+ },
275
+ "vision_tower.blocks.20.attn.qkv": {
276
+ "group_size": 64,
277
+ "bits": 8
278
+ },
279
+ "vision_tower.blocks.20.attn.proj": {
280
+ "group_size": 64,
281
+ "bits": 8
282
+ },
283
+ "vision_tower.blocks.20.mlp.linear_fc1": {
284
+ "group_size": 64,
285
+ "bits": 8
286
+ },
287
+ "vision_tower.blocks.21.attn.qkv": {
288
+ "group_size": 64,
289
+ "bits": 8
290
+ },
291
+ "vision_tower.blocks.21.attn.proj": {
292
+ "group_size": 64,
293
+ "bits": 8
294
+ },
295
+ "vision_tower.blocks.21.mlp.linear_fc1": {
296
+ "group_size": 64,
297
+ "bits": 8
298
+ },
299
+ "vision_tower.blocks.22.attn.qkv": {
300
+ "group_size": 64,
301
+ "bits": 8
302
+ },
303
+ "vision_tower.blocks.22.attn.proj": {
304
+ "group_size": 64,
305
+ "bits": 8
306
+ },
307
+ "vision_tower.blocks.22.mlp.linear_fc1": {
308
+ "group_size": 64,
309
+ "bits": 8
310
+ },
311
+ "vision_tower.blocks.23.attn.qkv": {
312
+ "group_size": 64,
313
+ "bits": 8
314
+ },
315
+ "vision_tower.blocks.23.attn.proj": {
316
+ "group_size": 64,
317
+ "bits": 8
318
+ },
319
+ "vision_tower.blocks.23.mlp.linear_fc1": {
320
+ "group_size": 64,
321
+ "bits": 8
322
+ },
323
+ "vision_tower.blocks.24.attn.qkv": {
324
+ "group_size": 64,
325
+ "bits": 8
326
+ },
327
+ "vision_tower.blocks.24.attn.proj": {
328
+ "group_size": 64,
329
+ "bits": 8
330
+ },
331
+ "vision_tower.blocks.24.mlp.linear_fc1": {
332
+ "group_size": 64,
333
+ "bits": 8
334
+ },
335
+ "vision_tower.blocks.25.attn.qkv": {
336
+ "group_size": 64,
337
+ "bits": 8
338
+ },
339
+ "vision_tower.blocks.25.attn.proj": {
340
+ "group_size": 64,
341
+ "bits": 8
342
+ },
343
+ "vision_tower.blocks.25.mlp.linear_fc1": {
344
+ "group_size": 64,
345
+ "bits": 8
346
+ },
347
+ "vision_tower.blocks.26.attn.qkv": {
348
+ "group_size": 64,
349
+ "bits": 8
350
+ },
351
+ "vision_tower.blocks.26.attn.proj": {
352
+ "group_size": 64,
353
+ "bits": 8
354
+ },
355
+ "vision_tower.blocks.26.mlp.linear_fc1": {
356
+ "group_size": 64,
357
+ "bits": 8
358
+ },
359
+ "vision_tower.merger.linear_fc1": {
360
+ "group_size": 64,
361
+ "bits": 8
362
+ },
363
+ "vision_tower.merger.linear_fc2": {
364
+ "group_size": 64,
365
+ "bits": 8
366
+ },
367
  "language_model.model.embed_tokens": {
368
  "bits": 4,
369
  "group_size": 64,
 
2859
  "group_size": 64,
2860
  "bits": 4,
2861
  "mode": "affine",
2862
+ "vision_tower.pos_embed": {
2863
+ "group_size": 64,
2864
+ "bits": 8
2865
+ },
2866
+ "vision_tower.blocks.0.attn.qkv": {
2867
+ "group_size": 64,
2868
+ "bits": 8
2869
+ },
2870
+ "vision_tower.blocks.0.attn.proj": {
2871
+ "group_size": 64,
2872
+ "bits": 8
2873
+ },
2874
+ "vision_tower.blocks.0.mlp.linear_fc1": {
2875
+ "group_size": 64,
2876
+ "bits": 8
2877
+ },
2878
+ "vision_tower.blocks.1.attn.qkv": {
2879
+ "group_size": 64,
2880
+ "bits": 8
2881
+ },
2882
+ "vision_tower.blocks.1.attn.proj": {
2883
+ "group_size": 64,
2884
+ "bits": 8
2885
+ },
2886
+ "vision_tower.blocks.1.mlp.linear_fc1": {
2887
+ "group_size": 64,
2888
+ "bits": 8
2889
+ },
2890
+ "vision_tower.blocks.2.attn.qkv": {
2891
+ "group_size": 64,
2892
+ "bits": 8
2893
+ },
2894
+ "vision_tower.blocks.2.attn.proj": {
2895
+ "group_size": 64,
2896
+ "bits": 8
2897
+ },
2898
+ "vision_tower.blocks.2.mlp.linear_fc1": {
2899
+ "group_size": 64,
2900
+ "bits": 8
2901
+ },
2902
+ "vision_tower.blocks.3.attn.qkv": {
2903
+ "group_size": 64,
2904
+ "bits": 8
2905
+ },
2906
+ "vision_tower.blocks.3.attn.proj": {
2907
+ "group_size": 64,
2908
+ "bits": 8
2909
+ },
2910
+ "vision_tower.blocks.3.mlp.linear_fc1": {
2911
+ "group_size": 64,
2912
+ "bits": 8
2913
+ },
2914
+ "vision_tower.blocks.4.attn.qkv": {
2915
+ "group_size": 64,
2916
+ "bits": 8
2917
+ },
2918
+ "vision_tower.blocks.4.attn.proj": {
2919
+ "group_size": 64,
2920
+ "bits": 8
2921
+ },
2922
+ "vision_tower.blocks.4.mlp.linear_fc1": {
2923
+ "group_size": 64,
2924
+ "bits": 8
2925
+ },
2926
+ "vision_tower.blocks.5.attn.qkv": {
2927
+ "group_size": 64,
2928
+ "bits": 8
2929
+ },
2930
+ "vision_tower.blocks.5.attn.proj": {
2931
+ "group_size": 64,
2932
+ "bits": 8
2933
+ },
2934
+ "vision_tower.blocks.5.mlp.linear_fc1": {
2935
+ "group_size": 64,
2936
+ "bits": 8
2937
+ },
2938
+ "vision_tower.blocks.6.attn.qkv": {
2939
+ "group_size": 64,
2940
+ "bits": 8
2941
+ },
2942
+ "vision_tower.blocks.6.attn.proj": {
2943
+ "group_size": 64,
2944
+ "bits": 8
2945
+ },
2946
+ "vision_tower.blocks.6.mlp.linear_fc1": {
2947
+ "group_size": 64,
2948
+ "bits": 8
2949
+ },
2950
+ "vision_tower.blocks.7.attn.qkv": {
2951
+ "group_size": 64,
2952
+ "bits": 8
2953
+ },
2954
+ "vision_tower.blocks.7.attn.proj": {
2955
+ "group_size": 64,
2956
+ "bits": 8
2957
+ },
2958
+ "vision_tower.blocks.7.mlp.linear_fc1": {
2959
+ "group_size": 64,
2960
+ "bits": 8
2961
+ },
2962
+ "vision_tower.blocks.8.attn.qkv": {
2963
+ "group_size": 64,
2964
+ "bits": 8
2965
+ },
2966
+ "vision_tower.blocks.8.attn.proj": {
2967
+ "group_size": 64,
2968
+ "bits": 8
2969
+ },
2970
+ "vision_tower.blocks.8.mlp.linear_fc1": {
2971
+ "group_size": 64,
2972
+ "bits": 8
2973
+ },
2974
+ "vision_tower.blocks.9.attn.qkv": {
2975
+ "group_size": 64,
2976
+ "bits": 8
2977
+ },
2978
+ "vision_tower.blocks.9.attn.proj": {
2979
+ "group_size": 64,
2980
+ "bits": 8
2981
+ },
2982
+ "vision_tower.blocks.9.mlp.linear_fc1": {
2983
+ "group_size": 64,
2984
+ "bits": 8
2985
+ },
2986
+ "vision_tower.blocks.10.attn.qkv": {
2987
+ "group_size": 64,
2988
+ "bits": 8
2989
+ },
2990
+ "vision_tower.blocks.10.attn.proj": {
2991
+ "group_size": 64,
2992
+ "bits": 8
2993
+ },
2994
+ "vision_tower.blocks.10.mlp.linear_fc1": {
2995
+ "group_size": 64,
2996
+ "bits": 8
2997
+ },
2998
+ "vision_tower.blocks.11.attn.qkv": {
2999
+ "group_size": 64,
3000
+ "bits": 8
3001
+ },
3002
+ "vision_tower.blocks.11.attn.proj": {
3003
+ "group_size": 64,
3004
+ "bits": 8
3005
+ },
3006
+ "vision_tower.blocks.11.mlp.linear_fc1": {
3007
+ "group_size": 64,
3008
+ "bits": 8
3009
+ },
3010
+ "vision_tower.blocks.12.attn.qkv": {
3011
+ "group_size": 64,
3012
+ "bits": 8
3013
+ },
3014
+ "vision_tower.blocks.12.attn.proj": {
3015
+ "group_size": 64,
3016
+ "bits": 8
3017
+ },
3018
+ "vision_tower.blocks.12.mlp.linear_fc1": {
3019
+ "group_size": 64,
3020
+ "bits": 8
3021
+ },
3022
+ "vision_tower.blocks.13.attn.qkv": {
3023
+ "group_size": 64,
3024
+ "bits": 8
3025
+ },
3026
+ "vision_tower.blocks.13.attn.proj": {
3027
+ "group_size": 64,
3028
+ "bits": 8
3029
+ },
3030
+ "vision_tower.blocks.13.mlp.linear_fc1": {
3031
+ "group_size": 64,
3032
+ "bits": 8
3033
+ },
3034
+ "vision_tower.blocks.14.attn.qkv": {
3035
+ "group_size": 64,
3036
+ "bits": 8
3037
+ },
3038
+ "vision_tower.blocks.14.attn.proj": {
3039
+ "group_size": 64,
3040
+ "bits": 8
3041
+ },
3042
+ "vision_tower.blocks.14.mlp.linear_fc1": {
3043
+ "group_size": 64,
3044
+ "bits": 8
3045
+ },
3046
+ "vision_tower.blocks.15.attn.qkv": {
3047
+ "group_size": 64,
3048
+ "bits": 8
3049
+ },
3050
+ "vision_tower.blocks.15.attn.proj": {
3051
+ "group_size": 64,
3052
+ "bits": 8
3053
+ },
3054
+ "vision_tower.blocks.15.mlp.linear_fc1": {
3055
+ "group_size": 64,
3056
+ "bits": 8
3057
+ },
3058
+ "vision_tower.blocks.16.attn.qkv": {
3059
+ "group_size": 64,
3060
+ "bits": 8
3061
+ },
3062
+ "vision_tower.blocks.16.attn.proj": {
3063
+ "group_size": 64,
3064
+ "bits": 8
3065
+ },
3066
+ "vision_tower.blocks.16.mlp.linear_fc1": {
3067
+ "group_size": 64,
3068
+ "bits": 8
3069
+ },
3070
+ "vision_tower.blocks.17.attn.qkv": {
3071
+ "group_size": 64,
3072
+ "bits": 8
3073
+ },
3074
+ "vision_tower.blocks.17.attn.proj": {
3075
+ "group_size": 64,
3076
+ "bits": 8
3077
+ },
3078
+ "vision_tower.blocks.17.mlp.linear_fc1": {
3079
+ "group_size": 64,
3080
+ "bits": 8
3081
+ },
3082
+ "vision_tower.blocks.18.attn.qkv": {
3083
+ "group_size": 64,
3084
+ "bits": 8
3085
+ },
3086
+ "vision_tower.blocks.18.attn.proj": {
3087
+ "group_size": 64,
3088
+ "bits": 8
3089
+ },
3090
+ "vision_tower.blocks.18.mlp.linear_fc1": {
3091
+ "group_size": 64,
3092
+ "bits": 8
3093
+ },
3094
+ "vision_tower.blocks.19.attn.qkv": {
3095
+ "group_size": 64,
3096
+ "bits": 8
3097
+ },
3098
+ "vision_tower.blocks.19.attn.proj": {
3099
+ "group_size": 64,
3100
+ "bits": 8
3101
+ },
3102
+ "vision_tower.blocks.19.mlp.linear_fc1": {
3103
+ "group_size": 64,
3104
+ "bits": 8
3105
+ },
3106
+ "vision_tower.blocks.20.attn.qkv": {
3107
+ "group_size": 64,
3108
+ "bits": 8
3109
+ },
3110
+ "vision_tower.blocks.20.attn.proj": {
3111
+ "group_size": 64,
3112
+ "bits": 8
3113
+ },
3114
+ "vision_tower.blocks.20.mlp.linear_fc1": {
3115
+ "group_size": 64,
3116
+ "bits": 8
3117
+ },
3118
+ "vision_tower.blocks.21.attn.qkv": {
3119
+ "group_size": 64,
3120
+ "bits": 8
3121
+ },
3122
+ "vision_tower.blocks.21.attn.proj": {
3123
+ "group_size": 64,
3124
+ "bits": 8
3125
+ },
3126
+ "vision_tower.blocks.21.mlp.linear_fc1": {
3127
+ "group_size": 64,
3128
+ "bits": 8
3129
+ },
3130
+ "vision_tower.blocks.22.attn.qkv": {
3131
+ "group_size": 64,
3132
+ "bits": 8
3133
+ },
3134
+ "vision_tower.blocks.22.attn.proj": {
3135
+ "group_size": 64,
3136
+ "bits": 8
3137
+ },
3138
+ "vision_tower.blocks.22.mlp.linear_fc1": {
3139
+ "group_size": 64,
3140
+ "bits": 8
3141
+ },
3142
+ "vision_tower.blocks.23.attn.qkv": {
3143
+ "group_size": 64,
3144
+ "bits": 8
3145
+ },
3146
+ "vision_tower.blocks.23.attn.proj": {
3147
+ "group_size": 64,
3148
+ "bits": 8
3149
+ },
3150
+ "vision_tower.blocks.23.mlp.linear_fc1": {
3151
+ "group_size": 64,
3152
+ "bits": 8
3153
+ },
3154
+ "vision_tower.blocks.24.attn.qkv": {
3155
+ "group_size": 64,
3156
+ "bits": 8
3157
+ },
3158
+ "vision_tower.blocks.24.attn.proj": {
3159
+ "group_size": 64,
3160
+ "bits": 8
3161
+ },
3162
+ "vision_tower.blocks.24.mlp.linear_fc1": {
3163
+ "group_size": 64,
3164
+ "bits": 8
3165
+ },
3166
+ "vision_tower.blocks.25.attn.qkv": {
3167
+ "group_size": 64,
3168
+ "bits": 8
3169
+ },
3170
+ "vision_tower.blocks.25.attn.proj": {
3171
+ "group_size": 64,
3172
+ "bits": 8
3173
+ },
3174
+ "vision_tower.blocks.25.mlp.linear_fc1": {
3175
+ "group_size": 64,
3176
+ "bits": 8
3177
+ },
3178
+ "vision_tower.blocks.26.attn.qkv": {
3179
+ "group_size": 64,
3180
+ "bits": 8
3181
+ },
3182
+ "vision_tower.blocks.26.attn.proj": {
3183
+ "group_size": 64,
3184
+ "bits": 8
3185
+ },
3186
+ "vision_tower.blocks.26.mlp.linear_fc1": {
3187
+ "group_size": 64,
3188
+ "bits": 8
3189
+ },
3190
+ "vision_tower.merger.linear_fc1": {
3191
+ "group_size": 64,
3192
+ "bits": 8
3193
+ },
3194
+ "vision_tower.merger.linear_fc2": {
3195
+ "group_size": 64,
3196
+ "bits": 8
3197
+ },
3198
  "language_model.model.embed_tokens": {
3199
  "bits": 4,
3200
  "group_size": 64,
 
5686
  "mode": "affine"
5687
  }
5688
  },
5689
+ "temperature": 1.0,
5690
  "text_config": {
5691
  "attention_bias": false,
5692
  "attention_dropout": 0.0,
 
5792
  ],
5793
  "partial_rotary_factor": 0.25,
5794
  "rope_theta": 10000000,
5795
+ "rope_type": "default"
5796
  },
5797
  "tie_word_embeddings": false,
5798
  "use_cache": true,
5799
  "vocab_size": 248320
5800
  },
5801
  "tie_word_embeddings": false,
5802
+ "top_k": 20,
5803
+ "top_p": 0.95,
5804
  "transformers_version": "5.7.0",
5805
  "video_token_id": 248057,
5806
+ "vision_config": {
5807
+ "deepstack_visual_indexes": [],
5808
+ "depth": 27,
5809
+ "dtype": "bfloat16",
5810
+ "hidden_act": "gelu_pytorch_tanh",
5811
+ "hidden_size": 1152,
5812
+ "in_channels": 3,
5813
+ "initializer_range": 0.02,
5814
+ "intermediate_size": 4304,
5815
+ "model_type": "qwen3_5_vision",
5816
+ "num_heads": 16,
5817
+ "num_position_embeddings": 2304,
5818
+ "out_hidden_size": 5120,
5819
+ "patch_size": 16,
5820
+ "spatial_merge_size": 2,
5821
+ "temporal_patch_size": 2
5822
+ },
5823
  "vision_end_token_id": 248054,
5824
  "vision_start_token_id": 248053
5825
  }
configuration.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"framework":"Pytorch","task":"image-text-to-text"}
merge_verify_report.json ADDED
@@ -0,0 +1,2225 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adapter_config": {
3
+ "alora_invocation_tokens": null,
4
+ "alpha_pattern": {},
5
+ "arrow_config": null,
6
+ "auto_mapping": null,
7
+ "base_model_name_or_path": "Qwen/Qwen3.6-27B",
8
+ "bias": "none",
9
+ "corda_config": null,
10
+ "ensure_weight_tying": false,
11
+ "eva_config": null,
12
+ "exclude_modules": null,
13
+ "fan_in_fan_out": false,
14
+ "inference_mode": true,
15
+ "init_lora_weights": true,
16
+ "layer_replication": null,
17
+ "layers_pattern": null,
18
+ "layers_to_transform": null,
19
+ "loftq_config": {},
20
+ "lora_alpha": 128,
21
+ "lora_bias": false,
22
+ "lora_dropout": 0.05,
23
+ "lora_ga_config": null,
24
+ "megatron_config": null,
25
+ "megatron_core": "megatron.core",
26
+ "modules_to_save": null,
27
+ "peft_type": "LORA",
28
+ "peft_version": "0.19.1",
29
+ "qalora_group_size": 16,
30
+ "r": 64,
31
+ "rank_pattern": {},
32
+ "revision": null,
33
+ "target_modules": "(?!.*visual).*\\.(q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)",
34
+ "target_parameters": null,
35
+ "task_type": "CAUSAL_LM",
36
+ "trainable_token_indices": null,
37
+ "use_bdlora": null,
38
+ "use_dora": false,
39
+ "use_qalora": false,
40
+ "use_rslora": false
41
+ },
42
+ "hf": {
43
+ "timings": {
44
+ "hf_load_base": 32.24131433299044,
45
+ "hf_apply_adapter": 9.03372309403494,
46
+ "hf_decode_adapter": 111.14524953003274,
47
+ "hf_logits_adapter": 1.4711488810135052,
48
+ "hf_merge": 0.42533229797845706,
49
+ "hf_save_merged": 132.39111311297165,
50
+ "hf_restore_base_only": 1.0795407029800117,
51
+ "hf_load_merged": 15.514058025029954,
52
+ "hf_decode_merged": 55.286309406976216,
53
+ "hf_logits_merged": 1.0814605370396748
54
+ },
55
+ "hf_adapter_vs_merged": {
56
+ "all_token_ids_exact_match": false,
57
+ "logit_max_abs_diff": 6.109375,
58
+ "logit_mean_abs_diff": 0.033503563340466246,
59
+ "argmax_agreement": 0.9929140832595217,
60
+ "per_prompt": [
61
+ {
62
+ "prompt": "What is the capital of France, and why is it historically significant?",
63
+ "token_ids_exact_match": false,
64
+ "first_divergence_idx": 20,
65
+ "logit_max_abs_diff": 6.109375,
66
+ "logit_mean_abs_diff": 0.03354217857122421,
67
+ "argmax_agreement": 0.9857142567634583
68
+ },
69
+ {
70
+ "prompt": "Solve step by step: a train travels 120 km in 1.5 hours. What is its average speed?",
71
+ "token_ids_exact_match": false,
72
+ "first_divergence_idx": 138,
73
+ "logit_max_abs_diff": 2.8125,
74
+ "logit_mean_abs_diff": 0.03629594296216965,
75
+ "argmax_agreement": 0.9931740760803223
76
+ },
77
+ {
78
+ "prompt": "Write a short Python function that returns the n-th Fibonacci number.",
79
+ "token_ids_exact_match": true,
80
+ "first_divergence_idx": null,
81
+ "logit_max_abs_diff": 2.78125,
82
+ "logit_mean_abs_diff": 0.03149561583995819,
83
+ "argmax_agreement": 0.9964157938957214
84
+ },
85
+ {
86
+ "prompt": "Explain in two sentences why the sky appears blue.",
87
+ "token_ids_exact_match": true,
88
+ "first_divergence_idx": null,
89
+ "logit_max_abs_diff": 3.09375,
90
+ "logit_mean_abs_diff": 0.032533302903175354,
91
+ "argmax_agreement": 0.9963898658752441
92
+ }
93
+ ]
94
+ }
95
+ },
96
+ "weight_check": {
97
+ "scaling": 2.0,
98
+ "n_modules": 256,
99
+ "n_modules_total": 256,
100
+ "max_abs_diff_vs_fp32_ref": 0.00022973120212554932,
101
+ "max_rel_diff_vs_fp32_ref": 0.0038910505827516317,
102
+ "worst_module": "model.language_model.layers.47.self_attn.v_proj.weight",
103
+ "non_target_checked": [
104
+ [
105
+ "model.language_model.embed_tokens.weight",
106
+ true
107
+ ],
108
+ [
109
+ "model.visual.patch_embed.proj.bias",
110
+ true
111
+ ],
112
+ [
113
+ "model.visual.patch_embed.proj.weight",
114
+ true
115
+ ],
116
+ [
117
+ "model.visual.pos_embed.weight",
118
+ true
119
+ ],
120
+ [
121
+ "mtp.pre_fc_norm_embedding.weight",
122
+ true
123
+ ],
124
+ [
125
+ "lm_head.weight",
126
+ true
127
+ ],
128
+ [
129
+ "model.language_model.layers.0.input_layernorm.weight",
130
+ true
131
+ ],
132
+ [
133
+ "model.language_model.layers.0.linear_attn.norm.weight",
134
+ true
135
+ ]
136
+ ],
137
+ "non_target_identical": 8,
138
+ "bf16_ulp_rel": 0.00390625
139
+ },
140
+ "engines": {
141
+ "A": {
142
+ "engine": "HF base+LoRA",
143
+ "gens": [
144
+ [
145
+ 8160,
146
+ 579,
147
+ 264,
148
+ 7047,
149
+ 1817,
150
+ 25,
151
+ 271,
152
+ 16,
153
+ 13,
154
+ 220,
155
+ 2972,
156
+ 27382,
157
+ 1386,
158
+ 279,
159
+ 6007,
160
+ 3296,
161
+ 92217,
162
+ 561,
163
+ 1156,
164
+ 16561,
165
+ 364,
166
+ 279,
167
+ 6511,
168
+ 314,
169
+ 9338,
170
+ 321,
171
+ 1141,
172
+ 13263,
173
+ 24556,
174
+ 13,
175
+ 198,
176
+ 17,
177
+ 13,
178
+ 220,
179
+ 2972,
180
+ 15666,
181
+ 279,
182
+ 1118,
183
+ 919,
184
+ 92217,
185
+ 561,
186
+ 6511,
187
+ 314,
188
+ 9338,
189
+ 369,
190
+ 11751,
191
+ 13,
192
+ 198,
193
+ 18,
194
+ 13,
195
+ 220,
196
+ 2972,
197
+ 4146,
198
+ 279,
199
+ 2018,
200
+ 919,
201
+ 92217,
202
+ 8169,
203
+ 369,
204
+ 11751,
205
+ 33649,
206
+ 4927,
207
+ 30,
208
+ 353,
209
+ 1144,
210
+ 310,
211
+ 3300,
212
+ 264,
213
+ 61446,
214
+ 694,
215
+ 15352,
216
+ 22527,
217
+ 314,
218
+ 1141,
219
+ 13263,
220
+ 12287,
221
+ 13,
222
+ 198,
223
+ 19,
224
+ 13,
225
+ 220,
226
+ 2972,
227
+ 1536,
228
+ 13263,
229
+ 3387,
230
+ 883,
231
+ 11751,
232
+ 92217,
233
+ 198,
234
+ 256,
235
+ 471,
236
+ 35725,
237
+ 14,
238
+ 39836,
239
+ 3712,
240
+ 25,
241
+ 75053,
242
+ 539,
243
+ 279,
244
+ 11751,
245
+ 3680,
246
+ 32459,
247
+ 11,
248
+ 2843,
249
+ 12386,
250
+ 3177,
251
+ 318,
252
+ 43,
253
+ 330,
254
+ 294,
255
+ 667,
256
+ 8,
257
+ 198,
258
+ 256,
259
+ 471,
260
+ 74509,
261
+ 4031,
262
+ 25,
263
+ 5651,
264
+ 314,
265
+ 6618,
266
+ 318,
267
+ 29187,
268
+ 314,
269
+ 11751,
270
+ 11,
271
+ 17508,
272
+ 3835,
273
+ 16,
274
+ 16,
275
+ 20,
276
+ 15,
277
+ 681,
278
+ 41679,
279
+ 17120,
280
+ 318,
281
+ 2536,
282
+ 265,
283
+ 9141,
284
+ 368,
285
+ 11,
286
+ 199949,
287
+ 28589,
288
+ 385,
289
+ 6502,
290
+ 681,
291
+ 4788,
292
+ 3990,
293
+ 314,
294
+ 279,
295
+ 8323,
296
+ 24284,
297
+ 198,
298
+ 256,
299
+ 471,
300
+ 52459,
301
+ 14,
302
+ 39836,
303
+ 17722,
304
+ 25,
305
+ 38901,
306
+ 321,
307
+ 19034,
308
+ 18156,
309
+ 11,
310
+ 8323,
311
+ 83079,
312
+ 579,
313
+ 10407,
314
+ 11,
315
+ 4257,
316
+ 314,
317
+ 18001,
318
+ 321,
319
+ 34508,
320
+ 198,
321
+ 256,
322
+ 471,
323
+ 220,
324
+ 16,
325
+ 23,
326
+ 337,
327
+ 9018,
328
+ 25,
329
+ 88664,
330
+ 3990,
331
+ 11,
332
+ 2660,
333
+ 314,
334
+ 279,
335
+ 8323,
336
+ 21440,
337
+ 318,
338
+ 16,
339
+ 22,
340
+ 23,
341
+ 24,
342
+ 681,
343
+ 43563,
344
+ 4369,
345
+ 11,
346
+ 78082,
347
+ 314,
348
+ 35925,
349
+ 11,
350
+ 3478,
351
+ 4788,
352
+ 92305,
353
+ 21090,
354
+ 198,
355
+ 256,
356
+ 471,
357
+ 220,
358
+ 16,
359
+ 24,
360
+ 337,
361
+ 9018,
362
+ 25,
363
+ 23770,
364
+ 1954,
365
+ 11,
366
+ 13697,
367
+ 1790,
368
+ 17039,
369
+ 579,
370
+ 74910,
371
+ 11,
372
+ 6590,
373
+ 2792,
374
+ 314,
375
+ 220,
376
+ 16,
377
+ 23,
378
+ 22,
379
+ 16,
380
+ 11,
381
+ 12387,
382
+ 94869,
383
+ 318,
384
+ 24415,
385
+ 10860,
386
+ 2074,
387
+ 11,
388
+ 16698,
389
+ 11,
390
+ 18550,
391
+ 8,
392
+ 198,
393
+ 256,
394
+ 471,
395
+ 220,
396
+ 17,
397
+ 15,
398
+ 337,
399
+ 9018,
400
+ 25
401
+ ],
402
+ [
403
+ 8160,
404
+ 579,
405
+ 264,
406
+ 7047,
407
+ 1817,
408
+ 25,
409
+ 271,
410
+ 16,
411
+ 13,
412
+ 220,
413
+ 2972,
414
+ 2014,
415
+ 53983,
416
+ 2570,
417
+ 5396,
418
+ 64700,
419
+ 198,
420
+ 256,
421
+ 471,
422
+ 21397,
423
+ 25,
424
+ 357,
425
+ 5257,
426
+ 33060,
427
+ 220,
428
+ 16,
429
+ 17,
430
+ 15,
431
+ 12754,
432
+ 303,
433
+ 220,
434
+ 16,
435
+ 13,
436
+ 20,
437
+ 3979,
438
+ 13,
439
+ 198,
440
+ 256,
441
+ 471,
442
+ 15380,
443
+ 25,
444
+ 3437,
445
+ 369,
446
+ 1141,
447
+ 5289,
448
+ 4478,
449
+ 30,
450
+ 198,
451
+ 256,
452
+ 471,
453
+ 5952,
454
+ 25,
455
+ 61148,
456
+ 2923,
457
+ 539,
458
+ 2923,
459
+ 13,
460
+ 271,
461
+ 17,
462
+ 13,
463
+ 220,
464
+ 2972,
465
+ 27382,
466
+ 1386,
467
+ 5141,
468
+ 73182,
469
+ 64700,
470
+ 198,
471
+ 256,
472
+ 471,
473
+ 22830,
474
+ 4478,
475
+ 14377,
476
+ 25,
477
+ 14739,
478
+ 283,
479
+ 30116,
480
+ 593,
481
+ 3984,
482
+ 198,
483
+ 256,
484
+ 471,
485
+ 15767,
486
+ 25,
487
+ 30116,
488
+ 283,
489
+ 220,
490
+ 16,
491
+ 17,
492
+ 15,
493
+ 12754,
494
+ 11,
495
+ 3984,
496
+ 283,
497
+ 220,
498
+ 16,
499
+ 13,
500
+ 20,
501
+ 3979,
502
+ 198,
503
+ 256,
504
+ 471,
505
+ 14235,
506
+ 310,
507
+ 10724,
508
+ 25,
509
+ 14739,
510
+ 303,
511
+ 12754,
512
+ 7294,
513
+ 271,
514
+ 18,
515
+ 13,
516
+ 220,
517
+ 2972,
518
+ 8046,
519
+ 13909,
520
+ 12,
521
+ 8046,
522
+ 71690,
523
+ 64700,
524
+ 198,
525
+ 256,
526
+ 471,
527
+ 14392,
528
+ 220,
529
+ 16,
530
+ 25,
531
+ 9357,
532
+ 1441,
533
+ 279,
534
+ 14377,
535
+ 364,
536
+ 5289,
537
+ 4478,
538
+ 13,
539
+ 198,
540
+ 257,
541
+ 14739,
542
+ 283,
543
+ 30116,
544
+ 1624,
545
+ 115,
546
+ 3984,
547
+ 198,
548
+ 256,
549
+ 471,
550
+ 14392,
551
+ 220,
552
+ 17,
553
+ 25,
554
+ 51032,
555
+ 303,
556
+ 279,
557
+ 2574,
558
+ 2663,
559
+ 13,
560
+ 198,
561
+ 257,
562
+ 14739,
563
+ 283,
564
+ 220,
565
+ 16,
566
+ 17,
567
+ 15,
568
+ 12754,
569
+ 1624,
570
+ 115,
571
+ 220,
572
+ 16,
573
+ 13,
574
+ 20,
575
+ 3979,
576
+ 198,
577
+ 256,
578
+ 471,
579
+ 14392,
580
+ 220,
581
+ 18,
582
+ 25,
583
+ 24209,
584
+ 279,
585
+ 12436,
586
+ 13,
587
+ 198,
588
+ 257,
589
+ 220,
590
+ 16,
591
+ 17,
592
+ 15,
593
+ 1624,
594
+ 115,
595
+ 220,
596
+ 16,
597
+ 13,
598
+ 20,
599
+ 283,
600
+ 220,
601
+ 23,
602
+ 15,
603
+ 198,
604
+ 256,
605
+ 471,
606
+ 14392,
607
+ 220,
608
+ 19,
609
+ 25,
610
+ 3130,
611
+ 279,
612
+ 1534,
613
+ 4087,
614
+ 440,
615
+ 7896,
616
+ 13,
617
+ 198,
618
+ 257,
619
+ 22830,
620
+ 4478,
621
+ 283,
622
+ 220,
623
+ 23,
624
+ 15,
625
+ 12754,
626
+ 7294,
627
+ 271,
628
+ 19,
629
+ 13,
630
+ 220,
631
+ 2972,
632
+ 31547,
633
+ 64700,
634
+ 198,
635
+ 256,
636
+ 471,
637
+ 4109,
638
+ 25,
639
+ 220,
640
+ 23,
641
+ 15,
642
+ 12754,
643
+ 7294,
644
+ 23985,
645
+ 220,
646
+ 16,
647
+ 13,
648
+ 20,
649
+ 304,
650
+ 283,
651
+ 220,
652
+ 16,
653
+ 17,
654
+ 15,
655
+ 12754,
656
+ 13,
657
+ 38643,
658
+ 13
659
+ ],
660
+ [
661
+ 8160,
662
+ 579,
663
+ 264,
664
+ 7047,
665
+ 1817,
666
+ 25,
667
+ 271,
668
+ 16,
669
+ 13,
670
+ 220,
671
+ 2972,
672
+ 15771,
673
+ 2598,
674
+ 2570,
675
+ 5952,
676
+ 92217,
677
+ 561,
678
+ 1156,
679
+ 6587,
680
+ 264,
681
+ 2716,
682
+ 12654,
683
+ 709,
684
+ 421,
685
+ 4523,
686
+ 279,
687
+ 307,
688
+ 7324,
689
+ 76938,
690
+ 1324,
691
+ 13,
692
+ 271,
693
+ 17,
694
+ 13,
695
+ 220,
696
+ 2972,
697
+ 27382,
698
+ 1386,
699
+ 5141,
700
+ 32666,
701
+ 92217,
702
+ 198,
703
+ 256,
704
+ 471,
705
+ 11106,
706
+ 25,
707
+ 12654,
708
+ 198,
709
+ 256,
710
+ 471,
711
+ 8984,
712
+ 25,
713
+ 307,
714
+ 7324,
715
+ 76938,
716
+ 1324,
717
+ 198,
718
+ 256,
719
+ 471,
720
+ 42794,
721
+ 25,
722
+ 10382,
723
+ 30211,
724
+ 66,
725
+ 1029,
726
+ 271,
727
+ 18,
728
+ 13,
729
+ 220,
730
+ 2972,
731
+ 35944,
732
+ 76938,
733
+ 19107,
734
+ 92217,
735
+ 198,
736
+ 256,
737
+ 471,
738
+ 426,
739
+ 7,
740
+ 15,
741
+ 8,
742
+ 283,
743
+ 220,
744
+ 15,
745
+ 198,
746
+ 256,
747
+ 471,
748
+ 426,
749
+ 7,
750
+ 16,
751
+ 8,
752
+ 283,
753
+ 220,
754
+ 16,
755
+ 198,
756
+ 256,
757
+ 471,
758
+ 426,
759
+ 1393,
760
+ 8,
761
+ 283,
762
+ 426,
763
+ 1393,
764
+ 12,
765
+ 16,
766
+ 8,
767
+ 478,
768
+ 426,
769
+ 1393,
770
+ 12,
771
+ 17,
772
+ 8,
773
+ 364,
774
+ 307,
775
+ 2518,
776
+ 220,
777
+ 17,
778
+ 198,
779
+ 256,
780
+ 471,
781
+ 6817,
782
+ 25,
783
+ 4188,
784
+ 16954,
785
+ 1151,
786
+ 440,
787
+ 426,
788
+ 7,
789
+ 16,
790
+ 11393,
791
+ 16,
792
+ 11,
793
+ 426,
794
+ 7,
795
+ 17,
796
+ 11393,
797
+ 16,
798
+ 11,
799
+ 694,
800
+ 279,
801
+ 5129,
802
+ 15019,
803
+ 20624,
804
+ 369,
805
+ 426,
806
+ 7,
807
+ 15,
808
+ 11393,
809
+ 15,
810
+ 11,
811
+ 426,
812
+ 7,
813
+ 16,
814
+ 11393,
815
+ 16,
816
+ 13,
817
+ 353,
818
+ 3172,
819
+ 8941,
820
+ 440,
821
+ 421,
822
+ 694,
823
+ 628,
824
+ 6088,
825
+ 424,
826
+ 413,
827
+ 4221,
828
+ 13,
829
+ 271,
830
+ 19,
831
+ 13,
832
+ 220,
833
+ 2972,
834
+ 79503,
835
+ 29806,
836
+ 14147,
837
+ 92217,
838
+ 198,
839
+ 256,
840
+ 471,
841
+ 348,
842
+ 75853,
843
+ 77340,
844
+ 8722,
845
+ 694,
846
+ 61240,
847
+ 318,
848
+ 46,
849
+ 7,
850
+ 17,
851
+ 83193,
852
+ 578,
853
+ 198,
854
+ 256,
855
+ 471,
856
+ 348,
857
+ 8275,
858
+ 1340,
859
+ 77340,
860
+ 85943,
861
+ 318,
862
+ 46,
863
+ 1393,
864
+ 8,
865
+ 854,
866
+ 11,
867
+ 496,
868
+ 7,
869
+ 16,
870
+ 8,
871
+ 3433,
872
+ 8,
873
+ 198,
874
+ 256,
875
+ 471,
876
+ 348,
877
+ 6479,
878
+ 26793,
879
+ 6880,
880
+ 77340,
881
+ 496,
882
+ 12171,
883
+ 307,
884
+ 8,
885
+ 694,
886
+ 777,
887
+ 6150,
888
+ 198,
889
+ 256,
890
+ 471,
891
+ 348,
892
+ 27852,
893
+ 294,
894
+ 579,
895
+ 14377,
896
+ 77340,
897
+ 496,
898
+ 7,
899
+ 16,
900
+ 8,
901
+ 694,
902
+ 18484,
903
+ 1406,
904
+ 15579,
905
+ 4562,
906
+ 198,
907
+ 256,
908
+ 471,
909
+ 348,
910
+ 44856,
911
+ 1954,
912
+ 77340,
913
+ 7441,
914
+ 694,
915
+ 11039,
916
+ 22373
917
+ ],
918
+ [
919
+ 8160,
920
+ 579,
921
+ 264,
922
+ 7047,
923
+ 1817,
924
+ 25,
925
+ 271,
926
+ 16,
927
+ 13,
928
+ 220,
929
+ 2972,
930
+ 2014,
931
+ 53983,
932
+ 2570,
933
+ 5396,
934
+ 64700,
935
+ 198,
936
+ 256,
937
+ 471,
938
+ 2972,
939
+ 14162,
940
+ 64700,
941
+ 8169,
942
+ 1503,
943
+ 279,
944
+ 12515,
945
+ 4833,
946
+ 6105,
947
+ 30,
948
+ 198,
949
+ 256,
950
+ 471,
951
+ 2972,
952
+ 17355,
953
+ 64700,
954
+ 79091,
955
+ 303,
956
+ 6681,
957
+ 1330,
958
+ 22157,
959
+ 13,
960
+ 271,
961
+ 17,
962
+ 13,
963
+ 220,
964
+ 2972,
965
+ 27382,
966
+ 1386,
967
+ 5141,
968
+ 35801,
969
+ 73182,
970
+ 64700,
971
+ 198,
972
+ 256,
973
+ 471,
974
+ 7806,
975
+ 4009,
976
+ 5435,
977
+ 660,
978
+ 7736,
979
+ 314,
980
+ 279,
981
+ 9155,
982
+ 19142,
983
+ 13,
984
+ 198,
985
+ 256,
986
+ 471,
987
+ 8964,
988
+ 579,
989
+ 16078,
990
+ 1100,
991
+ 9872,
992
+ 22602,
993
+ 89661,
994
+ 318,
995
+ 11855,
996
+ 5285,
997
+ 29899,
998
+ 8,
999
+ 777,
1000
+ 1056,
1001
+ 4860,
1002
+ 89661,
1003
+ 318,
1004
+ 1114,
1005
+ 25643,
1006
+ 5131,
1007
+ 8,
1008
+ 4016,
1009
+ 310,
1010
+ 12870,
1011
+ 60845,
1012
+ 69377,
1013
+ 13,
1014
+ 198,
1015
+ 256,
1016
+ 471,
1017
+ 5456,
1018
+ 6213,
1019
+ 513,
1020
+ 777,
1021
+ 15739,
1022
+ 310,
1023
+ 6105,
1024
+ 1056,
1025
+ 76990,
1026
+ 11,
1027
+ 748,
1028
+ 567,
1029
+ 42918,
1030
+ 279,
1031
+ 12515,
1032
+ 430,
1033
+ 6105,
1034
+ 13,
1035
+ 271,
1036
+ 18,
1037
+ 13,
1038
+ 220,
1039
+ 2972,
1040
+ 48401,
1041
+ 471,
1042
+ 76802,
1043
+ 220,
1044
+ 16,
1045
+ 318,
1046
+ 13423,
1047
+ 383,
1048
+ 69377,
1049
+ 188912,
1050
+ 198,
1051
+ 256,
1052
+ 7806,
1053
+ 4009,
1054
+ 5435,
1055
+ 660,
1056
+ 7736,
1057
+ 314,
1058
+ 279,
1059
+ 9155,
1060
+ 19142,
1061
+ 11,
1062
+ 694,
1063
+ 8964,
1064
+ 579,
1065
+ 16078,
1066
+ 1100,
1067
+ 9872,
1068
+ 22602,
1069
+ 6105,
1070
+ 89661,
1071
+ 777,
1072
+ 13057,
1073
+ 1056,
1074
+ 4860,
1075
+ 2438,
1076
+ 89661,
1077
+ 1472,
1078
+ 264,
1079
+ 1817,
1080
+ 2512,
1081
+ 12870,
1082
+ 60845,
1083
+ 69377,
1084
+ 13,
1085
+ 271,
1086
+ 19,
1087
+ 13,
1088
+ 220,
1089
+ 2972,
1090
+ 48401,
1091
+ 471,
1092
+ 76802,
1093
+ 220,
1094
+ 17,
1095
+ 318,
1096
+ 13423,
1097
+ 383,
1098
+ 19808,
1099
+ 188912,
1100
+ 198,
1101
+ 256,
1102
+ 10022,
1103
+ 76990,
1104
+ 3000,
1105
+ 1100,
1106
+ 9872,
1107
+ 1442,
1108
+ 777,
1109
+ 11,
1110
+ 1004,
1111
+ 6213,
1112
+ 513,
1113
+ 777,
1114
+ 15739,
1115
+ 310,
1116
+ 6105,
1117
+ 11,
1118
+ 864,
1119
+ 369,
1120
+ 3069,
1121
+ 567,
1122
+ 42918,
1123
+ 279,
1124
+ 12515,
1125
+ 430,
1126
+ 45239,
1127
+ 6105,
1128
+ 13,
1129
+ 271,
1130
+ 20,
1131
+ 13,
1132
+ 220,
1133
+ 2972,
1134
+ 3840,
1135
+ 84457,
1136
+ 64700,
1137
+ 198,
1138
+ 256,
1139
+ 471,
1140
+ 66172,
1141
+ 1330,
1142
+ 22157,
1143
+ 30,
1144
+ 7179,
1145
+ 13,
1146
+ 198,
1147
+ 256,
1148
+ 471,
1149
+ 17467,
1150
+ 1674,
1151
+ 3069,
1152
+ 279,
1153
+ 12515,
1154
+ 7701,
1155
+ 6105,
1156
+ 30,
1157
+ 7179,
1158
+ 13,
1159
+ 198,
1160
+ 256,
1161
+ 471,
1162
+ 35801,
1163
+ 722,
1164
+ 12995,
1165
+ 30,
1166
+ 7179,
1167
+ 13,
1168
+ 271,
1169
+ 21,
1170
+ 13,
1171
+ 220,
1172
+ 2972,
1173
+ 3812,
1174
+ 472
1175
+ ]
1176
+ ],
1177
+ "load_s": 41.27503742702538,
1178
+ "decode_s": 111.14524953003274,
1179
+ "gen_tokens": 1024
1180
+ },
1181
+ "B": {
1182
+ "engine": "HF merged",
1183
+ "gens": [
1184
+ [
1185
+ 8160,
1186
+ 579,
1187
+ 264,
1188
+ 7047,
1189
+ 1817,
1190
+ 25,
1191
+ 271,
1192
+ 16,
1193
+ 13,
1194
+ 220,
1195
+ 2972,
1196
+ 27382,
1197
+ 1386,
1198
+ 279,
1199
+ 6007,
1200
+ 3296,
1201
+ 92217,
1202
+ 561,
1203
+ 1156,
1204
+ 16561,
1205
+ 1330,
1206
+ 2433,
1207
+ 25,
1208
+ 198,
1209
+ 256,
1210
+ 471,
1211
+ 3437,
1212
+ 369,
1213
+ 279,
1214
+ 6511,
1215
+ 314,
1216
+ 9338,
1217
+ 30,
1218
+ 198,
1219
+ 256,
1220
+ 471,
1221
+ 8169,
1222
+ 369,
1223
+ 424,
1224
+ 33649,
1225
+ 4927,
1226
+ 30,
1227
+ 271,
1228
+ 17,
1229
+ 13,
1230
+ 220,
1231
+ 2972,
1232
+ 15666,
1233
+ 279,
1234
+ 1118,
1235
+ 919,
1236
+ 92217,
1237
+ 561,
1238
+ 6511,
1239
+ 314,
1240
+ 9338,
1241
+ 369,
1242
+ 11751,
1243
+ 13,
1244
+ 271,
1245
+ 18,
1246
+ 13,
1247
+ 220,
1248
+ 2972,
1249
+ 4146,
1250
+ 279,
1251
+ 2018,
1252
+ 919,
1253
+ 318,
1254
+ 42005,
1255
+ 908,
1256
+ 24556,
1257
+ 31230,
1258
+ 25,
1259
+ 353,
1260
+ 1144,
1261
+ 310,
1262
+ 3300,
1263
+ 264,
1264
+ 61446,
1265
+ 694,
1266
+ 15352,
1267
+ 22527,
1268
+ 314,
1269
+ 11751,
1270
+ 579,
1271
+ 13263,
1272
+ 24556,
1273
+ 13,
1274
+ 5141,
1275
+ 3387,
1276
+ 310,
1277
+ 3311,
1278
+ 25,
1279
+ 198,
1280
+ 256,
1281
+ 471,
1282
+ 35725,
1283
+ 14,
1284
+ 1993,
1285
+ 30839,
1286
+ 31304,
1287
+ 318,
1288
+ 43,
1289
+ 330,
1290
+ 294,
1291
+ 667,
1292
+ 11,
1293
+ 12386,
1294
+ 16576,
1295
+ 11,
1296
+ 39654,
1297
+ 6310,
1298
+ 8,
1299
+ 198,
1300
+ 256,
1301
+ 471,
1302
+ 29501,
1303
+ 2805,
1304
+ 42011,
1305
+ 3990,
1306
+ 314,
1307
+ 9338,
1308
+ 318,
1309
+ 3243,
1310
+ 278,
1311
+ 6511,
1312
+ 11,
1313
+ 83079,
1314
+ 11,
1315
+ 8323,
1316
+ 21440,
1317
+ 8,
1318
+ 198,
1319
+ 256,
1320
+ 471,
1321
+ 88664,
1322
+ 321,
1323
+ 19034,
1324
+ 18156,
1325
+ 318,
1326
+ 18642,
1327
+ 2305,
1328
+ 11,
1329
+ 58641,
1330
+ 11,
1331
+ 11991,
1332
+ 79835,
1333
+ 8,
1334
+ 198,
1335
+ 256,
1336
+ 471,
1337
+ 5001,
1338
+ 321,
1339
+ 7431,
1340
+ 318,
1341
+ 49,
1342
+ 6933,
1343
+ 23217,
1344
+ 11,
1345
+ 13977,
1346
+ 10860,
1347
+ 2074,
1348
+ 11,
1349
+ 6278,
1350
+ 1880,
1351
+ 11,
1352
+ 16698,
1353
+ 8,
1354
+ 198,
1355
+ 256,
1356
+ 471,
1357
+ 23269,
1358
+ 4038,
1359
+ 57902,
1360
+ 318,
1361
+ 36,
1362
+ 2992,
1363
+ 300,
1364
+ 21262,
1365
+ 11,
1366
+ 42027,
1367
+ 9141,
1368
+ 368,
1369
+ 11,
1370
+ 9440,
1371
+ 46890,
1372
+ 11,
1373
+ 4831,
1374
+ 5946,
1375
+ 198,
1376
+ 256,
1377
+ 471,
1378
+ 14955,
1379
+ 303,
1380
+ 3478,
1381
+ 13263,
1382
+ 4216,
1383
+ 318,
1384
+ 41772,
1385
+ 21440,
1386
+ 11,
1387
+ 4196,
1388
+ 14534,
1389
+ 11,
1390
+ 220,
1391
+ 16,
1392
+ 24,
1393
+ 21,
1394
+ 23,
1395
+ 21231,
1396
+ 11,
1397
+ 4831,
1398
+ 5946,
1399
+ 198,
1400
+ 256,
1401
+ 471,
1402
+ 7711,
1403
+ 9870,
1404
+ 318,
1405
+ 69,
1406
+ 8517,
1407
+ 11,
1408
+ 33847,
1409
+ 11,
1410
+ 59174,
1411
+ 11,
1412
+ 77989,
1413
+ 4196,
1414
+ 32043,
1415
+ 8,
1416
+ 271,
1417
+ 19,
1418
+ 13,
1419
+ 220,
1420
+ 2972,
1421
+ 22238,
1422
+ 279,
1423
+ 1965,
1424
+ 92217,
1425
+ 198,
1426
+ 256,
1427
+ 471,
1428
+ 6916,
1429
+ 4087,
1430
+ 1118,
1431
+ 198,
1432
+ 256,
1433
+ 471,
1434
+ 10446,
1435
+ 537,
1436
+ 13263,
1437
+ 24556,
1438
+ 1083,
1439
+ 2708,
1440
+ 11
1441
+ ],
1442
+ [
1443
+ 8160,
1444
+ 579,
1445
+ 264,
1446
+ 7047,
1447
+ 1817,
1448
+ 25,
1449
+ 271,
1450
+ 16,
1451
+ 13,
1452
+ 220,
1453
+ 2972,
1454
+ 2014,
1455
+ 53983,
1456
+ 2570,
1457
+ 5396,
1458
+ 64700,
1459
+ 198,
1460
+ 256,
1461
+ 471,
1462
+ 21397,
1463
+ 25,
1464
+ 357,
1465
+ 5257,
1466
+ 33060,
1467
+ 220,
1468
+ 16,
1469
+ 17,
1470
+ 15,
1471
+ 12754,
1472
+ 303,
1473
+ 220,
1474
+ 16,
1475
+ 13,
1476
+ 20,
1477
+ 3979,
1478
+ 13,
1479
+ 198,
1480
+ 256,
1481
+ 471,
1482
+ 15380,
1483
+ 25,
1484
+ 3437,
1485
+ 369,
1486
+ 1141,
1487
+ 5289,
1488
+ 4478,
1489
+ 30,
1490
+ 198,
1491
+ 256,
1492
+ 471,
1493
+ 5952,
1494
+ 25,
1495
+ 61148,
1496
+ 2923,
1497
+ 539,
1498
+ 2923,
1499
+ 13,
1500
+ 271,
1501
+ 17,
1502
+ 13,
1503
+ 220,
1504
+ 2972,
1505
+ 27382,
1506
+ 1386,
1507
+ 5141,
1508
+ 73182,
1509
+ 64700,
1510
+ 198,
1511
+ 256,
1512
+ 471,
1513
+ 22830,
1514
+ 4478,
1515
+ 14377,
1516
+ 25,
1517
+ 14739,
1518
+ 283,
1519
+ 30116,
1520
+ 593,
1521
+ 3984,
1522
+ 198,
1523
+ 256,
1524
+ 471,
1525
+ 15767,
1526
+ 25,
1527
+ 30116,
1528
+ 283,
1529
+ 220,
1530
+ 16,
1531
+ 17,
1532
+ 15,
1533
+ 12754,
1534
+ 11,
1535
+ 3984,
1536
+ 283,
1537
+ 220,
1538
+ 16,
1539
+ 13,
1540
+ 20,
1541
+ 3979,
1542
+ 198,
1543
+ 256,
1544
+ 471,
1545
+ 14235,
1546
+ 310,
1547
+ 10724,
1548
+ 25,
1549
+ 14739,
1550
+ 303,
1551
+ 12754,
1552
+ 7294,
1553
+ 271,
1554
+ 18,
1555
+ 13,
1556
+ 220,
1557
+ 2972,
1558
+ 8046,
1559
+ 13909,
1560
+ 12,
1561
+ 8046,
1562
+ 71690,
1563
+ 64700,
1564
+ 198,
1565
+ 256,
1566
+ 471,
1567
+ 14392,
1568
+ 220,
1569
+ 16,
1570
+ 25,
1571
+ 9357,
1572
+ 1441,
1573
+ 279,
1574
+ 14377,
1575
+ 364,
1576
+ 5289,
1577
+ 4478,
1578
+ 13,
1579
+ 198,
1580
+ 257,
1581
+ 471,
1582
+ 14739,
1583
+ 283,
1584
+ 30116,
1585
+ 1624,
1586
+ 115,
1587
+ 3984,
1588
+ 198,
1589
+ 256,
1590
+ 471,
1591
+ 14392,
1592
+ 220,
1593
+ 17,
1594
+ 25,
1595
+ 51032,
1596
+ 303,
1597
+ 279,
1598
+ 2574,
1599
+ 2663,
1600
+ 13,
1601
+ 198,
1602
+ 257,
1603
+ 471,
1604
+ 14739,
1605
+ 283,
1606
+ 220,
1607
+ 16,
1608
+ 17,
1609
+ 15,
1610
+ 12754,
1611
+ 1624,
1612
+ 115,
1613
+ 220,
1614
+ 16,
1615
+ 13,
1616
+ 20,
1617
+ 3979,
1618
+ 198,
1619
+ 256,
1620
+ 471,
1621
+ 14392,
1622
+ 220,
1623
+ 18,
1624
+ 25,
1625
+ 24209,
1626
+ 279,
1627
+ 12436,
1628
+ 13,
1629
+ 198,
1630
+ 257,
1631
+ 471,
1632
+ 220,
1633
+ 16,
1634
+ 17,
1635
+ 15,
1636
+ 1624,
1637
+ 115,
1638
+ 220,
1639
+ 16,
1640
+ 13,
1641
+ 20,
1642
+ 283,
1643
+ 220,
1644
+ 23,
1645
+ 15,
1646
+ 198,
1647
+ 256,
1648
+ 471,
1649
+ 14392,
1650
+ 220,
1651
+ 19,
1652
+ 25,
1653
+ 3130,
1654
+ 279,
1655
+ 1534,
1656
+ 4087,
1657
+ 440,
1658
+ 7896,
1659
+ 13,
1660
+ 198,
1661
+ 257,
1662
+ 471,
1663
+ 22830,
1664
+ 4478,
1665
+ 283,
1666
+ 220,
1667
+ 23,
1668
+ 15,
1669
+ 12754,
1670
+ 7294,
1671
+ 271,
1672
+ 19,
1673
+ 13,
1674
+ 220,
1675
+ 2972,
1676
+ 31547,
1677
+ 64700,
1678
+ 198,
1679
+ 256,
1680
+ 471,
1681
+ 4109,
1682
+ 25,
1683
+ 220,
1684
+ 23,
1685
+ 15,
1686
+ 12754,
1687
+ 7294,
1688
+ 23985,
1689
+ 220,
1690
+ 16,
1691
+ 13,
1692
+ 20,
1693
+ 304,
1694
+ 283,
1695
+ 220,
1696
+ 16,
1697
+ 17,
1698
+ 15
1699
+ ],
1700
+ [
1701
+ 8160,
1702
+ 579,
1703
+ 264,
1704
+ 7047,
1705
+ 1817,
1706
+ 25,
1707
+ 271,
1708
+ 16,
1709
+ 13,
1710
+ 220,
1711
+ 2972,
1712
+ 15771,
1713
+ 2598,
1714
+ 2570,
1715
+ 5952,
1716
+ 92217,
1717
+ 561,
1718
+ 1156,
1719
+ 6587,
1720
+ 264,
1721
+ 2716,
1722
+ 12654,
1723
+ 709,
1724
+ 421,
1725
+ 4523,
1726
+ 279,
1727
+ 307,
1728
+ 7324,
1729
+ 76938,
1730
+ 1324,
1731
+ 13,
1732
+ 271,
1733
+ 17,
1734
+ 13,
1735
+ 220,
1736
+ 2972,
1737
+ 27382,
1738
+ 1386,
1739
+ 5141,
1740
+ 32666,
1741
+ 92217,
1742
+ 198,
1743
+ 256,
1744
+ 471,
1745
+ 11106,
1746
+ 25,
1747
+ 12654,
1748
+ 198,
1749
+ 256,
1750
+ 471,
1751
+ 8984,
1752
+ 25,
1753
+ 307,
1754
+ 7324,
1755
+ 76938,
1756
+ 1324,
1757
+ 198,
1758
+ 256,
1759
+ 471,
1760
+ 42794,
1761
+ 25,
1762
+ 10382,
1763
+ 30211,
1764
+ 66,
1765
+ 1029,
1766
+ 271,
1767
+ 18,
1768
+ 13,
1769
+ 220,
1770
+ 2972,
1771
+ 35944,
1772
+ 76938,
1773
+ 19107,
1774
+ 92217,
1775
+ 198,
1776
+ 256,
1777
+ 471,
1778
+ 426,
1779
+ 7,
1780
+ 15,
1781
+ 8,
1782
+ 283,
1783
+ 220,
1784
+ 15,
1785
+ 198,
1786
+ 256,
1787
+ 471,
1788
+ 426,
1789
+ 7,
1790
+ 16,
1791
+ 8,
1792
+ 283,
1793
+ 220,
1794
+ 16,
1795
+ 198,
1796
+ 256,
1797
+ 471,
1798
+ 426,
1799
+ 1393,
1800
+ 8,
1801
+ 283,
1802
+ 426,
1803
+ 1393,
1804
+ 12,
1805
+ 16,
1806
+ 8,
1807
+ 478,
1808
+ 426,
1809
+ 1393,
1810
+ 12,
1811
+ 17,
1812
+ 8,
1813
+ 364,
1814
+ 307,
1815
+ 2518,
1816
+ 220,
1817
+ 17,
1818
+ 198,
1819
+ 256,
1820
+ 471,
1821
+ 6817,
1822
+ 25,
1823
+ 4188,
1824
+ 16954,
1825
+ 1151,
1826
+ 440,
1827
+ 426,
1828
+ 7,
1829
+ 16,
1830
+ 11393,
1831
+ 16,
1832
+ 11,
1833
+ 426,
1834
+ 7,
1835
+ 17,
1836
+ 11393,
1837
+ 16,
1838
+ 11,
1839
+ 694,
1840
+ 279,
1841
+ 5129,
1842
+ 15019,
1843
+ 20624,
1844
+ 369,
1845
+ 426,
1846
+ 7,
1847
+ 15,
1848
+ 11393,
1849
+ 15,
1850
+ 11,
1851
+ 426,
1852
+ 7,
1853
+ 16,
1854
+ 11393,
1855
+ 16,
1856
+ 13,
1857
+ 353,
1858
+ 3172,
1859
+ 8941,
1860
+ 440,
1861
+ 421,
1862
+ 694,
1863
+ 628,
1864
+ 6088,
1865
+ 424,
1866
+ 413,
1867
+ 4221,
1868
+ 13,
1869
+ 271,
1870
+ 19,
1871
+ 13,
1872
+ 220,
1873
+ 2972,
1874
+ 79503,
1875
+ 29806,
1876
+ 14147,
1877
+ 92217,
1878
+ 198,
1879
+ 256,
1880
+ 471,
1881
+ 348,
1882
+ 75853,
1883
+ 77340,
1884
+ 8722,
1885
+ 694,
1886
+ 61240,
1887
+ 318,
1888
+ 46,
1889
+ 7,
1890
+ 17,
1891
+ 83193,
1892
+ 578,
1893
+ 198,
1894
+ 256,
1895
+ 471,
1896
+ 348,
1897
+ 8275,
1898
+ 1340,
1899
+ 77340,
1900
+ 85943,
1901
+ 318,
1902
+ 46,
1903
+ 1393,
1904
+ 8,
1905
+ 854,
1906
+ 11,
1907
+ 496,
1908
+ 7,
1909
+ 16,
1910
+ 8,
1911
+ 3433,
1912
+ 8,
1913
+ 198,
1914
+ 256,
1915
+ 471,
1916
+ 348,
1917
+ 6479,
1918
+ 26793,
1919
+ 6880,
1920
+ 77340,
1921
+ 496,
1922
+ 12171,
1923
+ 307,
1924
+ 8,
1925
+ 694,
1926
+ 777,
1927
+ 6150,
1928
+ 198,
1929
+ 256,
1930
+ 471,
1931
+ 348,
1932
+ 27852,
1933
+ 294,
1934
+ 579,
1935
+ 14377,
1936
+ 77340,
1937
+ 496,
1938
+ 7,
1939
+ 16,
1940
+ 8,
1941
+ 694,
1942
+ 18484,
1943
+ 1406,
1944
+ 15579,
1945
+ 4562,
1946
+ 198,
1947
+ 256,
1948
+ 471,
1949
+ 348,
1950
+ 44856,
1951
+ 1954,
1952
+ 77340,
1953
+ 7441,
1954
+ 694,
1955
+ 11039,
1956
+ 22373
1957
+ ],
1958
+ [
1959
+ 8160,
1960
+ 579,
1961
+ 264,
1962
+ 7047,
1963
+ 1817,
1964
+ 25,
1965
+ 271,
1966
+ 16,
1967
+ 13,
1968
+ 220,
1969
+ 2972,
1970
+ 2014,
1971
+ 53983,
1972
+ 2570,
1973
+ 5396,
1974
+ 64700,
1975
+ 198,
1976
+ 256,
1977
+ 471,
1978
+ 2972,
1979
+ 14162,
1980
+ 64700,
1981
+ 8169,
1982
+ 1503,
1983
+ 279,
1984
+ 12515,
1985
+ 4833,
1986
+ 6105,
1987
+ 30,
1988
+ 198,
1989
+ 256,
1990
+ 471,
1991
+ 2972,
1992
+ 17355,
1993
+ 64700,
1994
+ 79091,
1995
+ 303,
1996
+ 6681,
1997
+ 1330,
1998
+ 22157,
1999
+ 13,
2000
+ 271,
2001
+ 17,
2002
+ 13,
2003
+ 220,
2004
+ 2972,
2005
+ 27382,
2006
+ 1386,
2007
+ 5141,
2008
+ 35801,
2009
+ 73182,
2010
+ 64700,
2011
+ 198,
2012
+ 256,
2013
+ 471,
2014
+ 7806,
2015
+ 4009,
2016
+ 5435,
2017
+ 660,
2018
+ 7736,
2019
+ 314,
2020
+ 279,
2021
+ 9155,
2022
+ 19142,
2023
+ 13,
2024
+ 198,
2025
+ 256,
2026
+ 471,
2027
+ 8964,
2028
+ 579,
2029
+ 16078,
2030
+ 1100,
2031
+ 9872,
2032
+ 22602,
2033
+ 89661,
2034
+ 318,
2035
+ 11855,
2036
+ 5285,
2037
+ 29899,
2038
+ 8,
2039
+ 777,
2040
+ 1056,
2041
+ 4860,
2042
+ 89661,
2043
+ 318,
2044
+ 1114,
2045
+ 25643,
2046
+ 5131,
2047
+ 8,
2048
+ 4016,
2049
+ 310,
2050
+ 12870,
2051
+ 60845,
2052
+ 69377,
2053
+ 13,
2054
+ 198,
2055
+ 256,
2056
+ 471,
2057
+ 5456,
2058
+ 6213,
2059
+ 513,
2060
+ 777,
2061
+ 15739,
2062
+ 310,
2063
+ 6105,
2064
+ 1056,
2065
+ 76990,
2066
+ 11,
2067
+ 748,
2068
+ 567,
2069
+ 42918,
2070
+ 279,
2071
+ 12515,
2072
+ 430,
2073
+ 6105,
2074
+ 13,
2075
+ 271,
2076
+ 18,
2077
+ 13,
2078
+ 220,
2079
+ 2972,
2080
+ 48401,
2081
+ 471,
2082
+ 76802,
2083
+ 220,
2084
+ 16,
2085
+ 318,
2086
+ 13423,
2087
+ 383,
2088
+ 69377,
2089
+ 188912,
2090
+ 198,
2091
+ 256,
2092
+ 7806,
2093
+ 4009,
2094
+ 5435,
2095
+ 660,
2096
+ 7736,
2097
+ 314,
2098
+ 279,
2099
+ 9155,
2100
+ 19142,
2101
+ 11,
2102
+ 694,
2103
+ 8964,
2104
+ 579,
2105
+ 16078,
2106
+ 1100,
2107
+ 9872,
2108
+ 22602,
2109
+ 6105,
2110
+ 89661,
2111
+ 777,
2112
+ 13057,
2113
+ 1056,
2114
+ 4860,
2115
+ 2438,
2116
+ 89661,
2117
+ 1472,
2118
+ 264,
2119
+ 1817,
2120
+ 2512,
2121
+ 12870,
2122
+ 60845,
2123
+ 69377,
2124
+ 13,
2125
+ 271,
2126
+ 19,
2127
+ 13,
2128
+ 220,
2129
+ 2972,
2130
+ 48401,
2131
+ 471,
2132
+ 76802,
2133
+ 220,
2134
+ 17,
2135
+ 318,
2136
+ 13423,
2137
+ 383,
2138
+ 19808,
2139
+ 188912,
2140
+ 198,
2141
+ 256,
2142
+ 10022,
2143
+ 76990,
2144
+ 3000,
2145
+ 1100,
2146
+ 9872,
2147
+ 1442,
2148
+ 777,
2149
+ 11,
2150
+ 1004,
2151
+ 6213,
2152
+ 513,
2153
+ 777,
2154
+ 15739,
2155
+ 310,
2156
+ 6105,
2157
+ 11,
2158
+ 864,
2159
+ 369,
2160
+ 3069,
2161
+ 567,
2162
+ 42918,
2163
+ 279,
2164
+ 12515,
2165
+ 430,
2166
+ 45239,
2167
+ 6105,
2168
+ 13,
2169
+ 271,
2170
+ 20,
2171
+ 13,
2172
+ 220,
2173
+ 2972,
2174
+ 3840,
2175
+ 84457,
2176
+ 64700,
2177
+ 198,
2178
+ 256,
2179
+ 471,
2180
+ 66172,
2181
+ 1330,
2182
+ 22157,
2183
+ 30,
2184
+ 7179,
2185
+ 13,
2186
+ 198,
2187
+ 256,
2188
+ 471,
2189
+ 17467,
2190
+ 1674,
2191
+ 3069,
2192
+ 279,
2193
+ 12515,
2194
+ 7701,
2195
+ 6105,
2196
+ 30,
2197
+ 7179,
2198
+ 13,
2199
+ 198,
2200
+ 256,
2201
+ 471,
2202
+ 35801,
2203
+ 722,
2204
+ 12995,
2205
+ 30,
2206
+ 7179,
2207
+ 13,
2208
+ 271,
2209
+ 21,
2210
+ 13,
2211
+ 220,
2212
+ 2972,
2213
+ 3812,
2214
+ 472
2215
+ ]
2216
+ ],
2217
+ "load_s": 15.514058025029954,
2218
+ "decode_s": 55.286309406976216,
2219
+ "gen_tokens": 1024
2220
+ }
2221
+ },
2222
+ "argmax_matrix": {
2223
+ "A_vs_B": 0.9937777777777778
2224
+ }
2225
+ }
model-00003-of-00003.safetensors → model-00001-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4512f7fe8ff107189a81333d69f2b209aae0f7adfbabbc3b37a3701aa2bc17dd
3
- size 4741362734
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61ed1acb84738d23889116120c2b803fa906484cd7d1c89888e93241bfa29c24
3
+ size 5319000533
model-00001-of-00003.safetensors → model-00002-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2282bc250064a52f0f3d1e21be3019141a133e6fa52b36e372e1d46f5a5dc3ab
3
- size 5359250448
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:37edcf14d5133b836ca85d3414094deeb2fde6e91da86dcb62ce8b2fbc061484
3
+ size 5367483332
model-00002-of-00003.safetensors → model-00003-of-00004.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d24ce95a70d3b76caa9c1e9b5d8c8227a603d2fd13fd93e15baff13a774d94c1
3
- size 5367974699
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f32b0a0cacb8fa6dd661926b7751d2dee26421d7e1b7a1ba7f8082e439187969
3
+ size 5359502926
model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f67ff545147a7bdea48bc0c9a5cb7ca73b569f667f86e8e9ffa922fff349515b
3
+ size 39731338
model.safetensors.index.json CHANGED
The diff for this file is too large to render. See raw diff
 
preprocessor_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "size": {
3
+ "longest_edge": 16777216,
4
+ "shortest_edge": 65536
5
+ },
6
+ "patch_size": 16,
7
+ "temporal_patch_size": 2,
8
+ "merge_size": 2,
9
+ "image_mean": [
10
+ 0.5,
11
+ 0.5,
12
+ 0.5
13
+ ],
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "processor_class": "Qwen3VLProcessor",
20
+ "image_processor_type": "Qwen2VLImageProcessorFast"
21
+ }
processor_config.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": true,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "image_mean": [
7
+ 0.5,
8
+ 0.5,
9
+ 0.5
10
+ ],
11
+ "image_processor_type": "Qwen3VLImageProcessor",
12
+ "image_std": [
13
+ 0.5,
14
+ 0.5,
15
+ 0.5
16
+ ],
17
+ "max_pixels": 16777216,
18
+ "merge_size": 2,
19
+ "min_pixels": 65536,
20
+ "patch_size": 16,
21
+ "rescale_factor": 0.00392156862745098,
22
+ "temporal_patch_size": 2
23
+ },
24
+ "processor_class": "Qwen3VLProcessor",
25
+ "video_processor": {
26
+ "do_convert_rgb": true,
27
+ "do_normalize": true,
28
+ "do_rescale": true,
29
+ "fps": 2.0,
30
+ "image_mean": [
31
+ 0.5,
32
+ 0.5,
33
+ 0.5
34
+ ],
35
+ "image_std": [
36
+ 0.5,
37
+ 0.5,
38
+ 0.5
39
+ ],
40
+ "max_frames": 768,
41
+ "max_pixels": 25165824,
42
+ "merge_size": 2,
43
+ "min_frames": 4,
44
+ "min_pixels": 4096,
45
+ "patch_size": 16,
46
+ "rescale_factor": 0.00392156862745098,
47
+ "temporal_patch_size": 2,
48
+ "video_processor_type": "Qwen3VLVideoProcessor"
49
+ }
50
+ }
tokenizer_config.json CHANGED
@@ -23,9 +23,9 @@
23
  },
24
  "pad_token": "<|endoftext|>",
25
  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
 
26
  "split_special_tokens": false,
27
  "tokenizer_class": "Qwen2Tokenizer",
28
- "tool_parser_type": "qwen3_coder",
29
  "unk_token": null,
30
  "video_token": "<|video_pad|>",
31
  "vision_bos_token": "<|vision_start|>",
 
23
  },
24
  "pad_token": "<|endoftext|>",
25
  "pretokenize_regex": "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
26
+ "processor_class": "Qwen3VLProcessor",
27
  "split_special_tokens": false,
28
  "tokenizer_class": "Qwen2Tokenizer",
 
29
  "unk_token": null,
30
  "video_token": "<|video_pad|>",
31
  "vision_bos_token": "<|vision_start|>",
video_preprocessor_config.json ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "size": {
3
+ "longest_edge": 25165824,
4
+ "shortest_edge": 4096
5
+ },
6
+ "patch_size": 16,
7
+ "temporal_patch_size": 2,
8
+ "merge_size": 2,
9
+ "image_mean": [
10
+ 0.5,
11
+ 0.5,
12
+ 0.5
13
+ ],
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "processor_class": "Qwen3VLProcessor",
20
+ "video_processor_type": "Qwen3VLVideoProcessor"
21
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff
 
weight_check_report.json ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "scaling": 2.0,
3
+ "n_modules": 256,
4
+ "n_bitwise_exact_bf16_replicate": 0,
5
+ "max_abs_diff_vs_fp32_ref": 0.00024040043354034424,
6
+ "max_rel_diff_vs_fp32_ref": 0.0038910505827516317,
7
+ "worst_module": "model.language_model.layers.47.self_attn.v_proj.weight",
8
+ "non_target_checked": [
9
+ {
10
+ "key": "model.language_model.embed_tokens.weight",
11
+ "identical": true
12
+ },
13
+ {
14
+ "key": "model.visual.patch_embed.proj.bias",
15
+ "identical": true
16
+ },
17
+ {
18
+ "key": "model.visual.patch_embed.proj.weight",
19
+ "identical": true
20
+ },
21
+ {
22
+ "key": "model.visual.pos_embed.weight",
23
+ "identical": true
24
+ },
25
+ {
26
+ "key": "lm_head.weight",
27
+ "identical": true
28
+ },
29
+ {
30
+ "key": "model.language_model.layers.0.input_layernorm.weight",
31
+ "identical": true
32
+ },
33
+ {
34
+ "key": "model.language_model.layers.0.linear_attn.norm.weight",
35
+ "identical": true
36
+ },
37
+ {
38
+ "key": "model.language_model.layers.0.post_attention_layernorm.weight",
39
+ "identical": true
40
+ }
41
+ ],
42
+ "non_target_identical": 8
43
+ }