Hugging Face's logo

CollectionStudio
/
Inkling-Small-NVFP4

Image-Text-to-Text
Transformers
Safetensors
inkling_mm_model
conversational
audio-text-to-text
Mixture of Experts
8-bit precision
Model card Files
xet
 Community
CollectionStudio aurick commited on
Commit
ffc03ae
·
0 Parent(s):

Duplicate from thinkingmachines/Inkling-Small-NVFP4

Browse files

Co-authored-by: Aurick Qiao <aurick@users.noreply.huggingface.co>

Files changed (21) hide show
  1. .gitattributes +36 -0
  2. README.md +790 -0
  3. chat_template.jinja +129 -0
  4. config.json +114 -0
  5. hf_quant_config.json +411 -0
  6. model-00001-of-00009.safetensors +3 -0
  7. model-00002-of-00009.safetensors +3 -0
  8. model-00003-of-00009.safetensors +3 -0
  9. model-00004-of-00009.safetensors +3 -0
  10. model-00005-of-00009.safetensors +3 -0
  11. model-00006-of-00009.safetensors +3 -0
  12. model-00007-of-00009.safetensors +3 -0
  13. model-00008-of-00009.safetensors +3 -0
  14. model-00009-of-00009.safetensors +3 -0
  15. model.safetensors.index.json +0 -0
  16. mtp.safetensors +3 -0
  17. processor_config.json +46 -0
  18. special_tokens_map.json +22 -0
  19. tiktoken/tokenizer.model +3 -0
  20. tokenizer.json +3 -0
  21. tokenizer_config.json +508 -0
.gitattributes ADDED
@@ -0,0 +1,36 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz filter=lfs diff=lfs merge=lfs -text
33
+ *.zip filter=lfs diff=lfs merge=lfs -text
34
+ *.zst filter=lfs diff=lfs merge=lfs -text
35
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,790 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ license_link: https://www.apache.org/licenses/LICENSE-2.0
4
+ pipeline_tag: image-text-to-text
5
+ tags:
6
+ - conversational
7
+ - image-text-to-text
8
+ - audio-text-to-text
9
+ - moe
10
+ library_name: transformers
11
+ base_model:
12
+ - thinkingmachines/Inkling-Small
13
+ base_model_relation: quantized
14
+ ---
15
+
16
+ # Inkling
17
+
18
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/630e8f0bf6f6d700f50ebd2e/AvmDwmrWRMnKjOWvmLieg.png" style="display: block;margin-left: auto;margin-right: auto;width: 30%;">
19
+
20
+ <p align="center">
21
+ <a href="https://huggingface.co/thinkingmachines/Inkling-Small">BF16</a> |
22
+ <a href="https://huggingface.co/thinkingmachines/Inkling-Small-NVFP4">NVFP4</a> |
23
+ <a href="https://tinker.thinkingmachines.ai/playground">Playground</a> |
24
+ <a href="https://github.com/thinking-machines-lab/tinker-cookbook">Tinker Cookbook</a> |
25
+ <a href="https://thinkingmachines.ai/model-acceptable-use-policy">Acceptable Use</a>
26
+ </p>
27
+
28
+ ## 1. General Information
29
+
30
+ Inkling-Small is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs. It is intended for use in English and other languages, and across multiple coding languages. The model is designed to be used by developers building AI-powered applications, including agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation systems, and is suitable for general-purpose conversational use, instruction-following, and other natural language and multimodal tasks. It is released with open weights to support research, fine-tuning and integration into third-party products by downstream developers.
31
+
32
+ **Languages:** English, with general multilingual capabilities across other languages.
33
+
34
+ ## 2. Getting Started
35
+
36
+ Try Inkling-Small on the [Tinker Playground](https://tinker.thinkingmachines.ai/playground) or access via API using the [Tinker Cookbook](https://github.com/thinking-machines-lab/tinker-cookbook).
37
+
38
+ Inkling-Small supports local deployment using the following open-source libraries:
39
+
40
+ * SGLang ([recipe](https://docs.sglang.io/cookbook/autoregressive/ThinkingMachines/Inkling-Small))
41
+ * vLLM ([recipe](https://recipes.vllm.ai/thinkingmachines/Inkling-Small))
42
+ * TokenSpeed ([recipe](https://lightseek.org/tokenspeed/recipes/models#Inkling))
43
+ * Unsloth ([recipe](https://unsloth.ai/docs/models/inkling))
44
+ * Huggingface ([recipe](https://hf.co/blog/thinkingmachines-inkling))
45
+
46
+ API access is also available through third party inference providers.
47
+
48
+ ## 3. Model Properties
49
+
50
+ ### Model type
51
+
52
+ Multimodal autoregressive transformer
53
+
54
+ ### Architecture type
55
+
56
+ A 42-layer decoder-only transformer with a sparse Mixture-of-Experts (MoE) feed-forward backbone: each token is routed to 6 of 256 experts, plus 2 shared experts active on every token. Attention is a hybrid of local and global layers. The model is natively multimodal — images are encoded via a hierarchical patch encoder, and audio via discrete token encoding — with all modalities projected into a shared hidden space and processed jointly by the decoder.
57
+
58
+ ### Parameters
59
+
60
+ 276B total, 12B active
61
+
62
+ ### Numerics support
63
+
64
+ BF16 and NVFP4
65
+
66
+ ### Input modalities
67
+
68
+ Inkling-Small accepts text, image, and audio inputs:
69
+
70
+ - Text: UTF-8 encoded text
71
+ - Image: Any pixel-based image input. For optimal performance, each image dimension should be between 40px to 4096px.
72
+ - Audio: WAV format, sampled at 16kHz. For optimal performance, audio length should ideally be under 2 mins.
73
+
74
+ ### Output modalities
75
+
76
+ Inkling-Small generates output as UTF-8 encoded text.
77
+
78
+ ## 4. Training
79
+
80
+ Training data includes a broad variety of content types, including text, images, audio, video.
81
+
82
+ Training data for the model was drawn from publicly available sources, acquired from third-parties, or synthetically generated or augmented. Publicly available data includes content from the public internet and publicly accessible repositories.
83
+
84
+ The training data curation process includes cleaning, processing, and modifying datasets. These processing steps, which vary by data type, may include deduplication and filtering to remove junk or other low-quality data, or to advance safety or other objectives.
85
+
86
+ ## 5. Evaluations
87
+
88
+ <table class="benchmark-results-table benchmark-results-categorized">
89
+ <colgroup>
90
+ <col class="benchmark-name-column">
91
+ <col class="benchmark-model-column" span="10">
92
+ </colgroup>
93
+ <thead>
94
+ <tr class="benchmark-model-group-row">
95
+ <td class="benchmark-heading-spacer" aria-hidden="true"></td>
96
+ <th class="benchmark-model-group" colspan="7" scope="colgroup">
97
+ <span class="benchmark-model-group-anchor"><span class="benchmark-model-group-label">Open weights</span></span>
98
+ </th>
99
+ <th class="benchmark-model-group" colspan="3" scope="colgroup">
100
+ <span class="benchmark-model-group-anchor"><span class="benchmark-model-group-label">Closed weights</span></span>
101
+ </th>
102
+ </tr>
103
+ <tr class="benchmark-model-header-row">
104
+ <td class="benchmark-heading-spacer" aria-hidden="true"></td>
105
+ <th class="benchmark-model benchmark-instant-start"><span class="benchmark-lock-content"><span class="benchmark-model-name">
106
+ <span class="benchmark-model-name-line">Inkling-Small</span></span></span></th>
107
+ <th class="benchmark-model"><span class="benchmark-model-name">
108
+ <span class="benchmark-model-name-line">Qwen3.5</span> <span class="benchmark-model-name-line">397B-A17B</span></span></th>
109
+ <th class="benchmark-model"><span class="benchmark-model-name">
110
+ <span class="benchmark-model-name-line">MiMo V2.5</span></span></th>
111
+ <th class="benchmark-model"><span class="benchmark-model-name">
112
+ <span class="benchmark-model-name-line">Minimax M2.7</span></span></th>
113
+ <th class="benchmark-model"><span class="benchmark-model-name">
114
+ <span class="benchmark-model-name-line">DeepSeek V4</span> <span class="benchmark-model-name-line">Flash</span></span></th>
115
+ <th class="benchmark-model"><span class="benchmark-model-name">
116
+ <span class="benchmark-model-name-line">Nemotron 3</span> <span class="benchmark-model-name-line">Ultra</span></span></th>
117
+ <th class="benchmark-model benchmark-instant-start"><span class="benchmark-lock-content"><span class="benchmark-model-name">
118
+ <span class="benchmark-model-name-line">Inkling</span></span></span></th>
119
+ <th class="benchmark-model"><span class="benchmark-model-name">
120
+ <span class="benchmark-model-name-line">Claude 4.5</span> <span class="benchmark-model-name-line">Haiku</span></span></th>
121
+ <th class="benchmark-model"><span class="benchmark-model-name">
122
+ <span class="benchmark-model-name-line">Gemini 3.5</span> <span class="benchmark-model-name-line">Flash-Lite</span></span></th>
123
+ <th class="benchmark-model"><span class="benchmark-model-name">
124
+ <span class="benchmark-model-name-line">GPT 5.6</span> <span class="benchmark-model-name-line">Luna</span></span></th>
125
+ </tr>
126
+ </thead>
127
+ <tbody class="benchmark-category-group">
128
+ <tr class="benchmark-category-row">
129
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Model Info</span></th>
130
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
131
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
132
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
133
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
134
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
135
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
136
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
137
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
138
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
139
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
140
+ </tr>
141
+ <tr>
142
+ <td class="benchmark-name">
143
+ <span class="benchmark-title">AA Index</span> <span class="benchmark-subtitle">(v4.1)</span>
144
+ </td>
145
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">40.0%</span></td>
146
+ <td class="benchmark-value">34.0%</td>
147
+ <td class="benchmark-value">37.0%</td>
148
+ <td class="benchmark-value">38.0%</td>
149
+ <td class="benchmark-value">40.0%</td>
150
+ <td class="benchmark-value">38.0%</td>
151
+ <td class="benchmark-value">41.0%</td>
152
+ <td class="benchmark-value">30.0%</td>
153
+ <td class="benchmark-value">36.0%</td>
154
+ <td class="benchmark-value">49.0%</td>
155
+ </tr>
156
+ <tr>
157
+ <td class="benchmark-name">
158
+ <span class="benchmark-title">Params (B)</span> <span class="benchmark-subtitle">(activated / total)</span>
159
+ </td>
160
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">12 / 276</span></td>
161
+ <td class="benchmark-value">17 / 397</td>
162
+ <td class="benchmark-value">15 / 310</td>
163
+ <td class="benchmark-value">10 / 230</td>
164
+ <td class="benchmark-value">13 / 284</td>
165
+ <td class="benchmark-value">55 / 550</td>
166
+ <td class="benchmark-value">41 / 975</td>
167
+ <td class="benchmark-value">–</td>
168
+ <td class="benchmark-value">–</td>
169
+ <td class="benchmark-value">–</td>
170
+ </tr>
171
+ </tbody>
172
+ <tbody class="benchmark-category-group">
173
+ <tr class="benchmark-category-row">
174
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Agentic (coding)</span></th>
175
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
176
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
177
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
178
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
179
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
180
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
181
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
182
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
183
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
184
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
185
+ </tr>
186
+ <tr>
187
+ <td class="benchmark-name">
188
+ <span class="benchmark-title">SWEBench Verified</span>
189
+ </td>
190
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">80.2%</span></td>
191
+ <td class="benchmark-value">76.4%</td>
192
+ <td class="benchmark-value">71.0%</td>
193
+ <td class="benchmark-value">79.9%</td>
194
+ <td class="benchmark-value">79.0%</td>
195
+ <td class="benchmark-value">70.7%</td>
196
+ <td class="benchmark-value">77.6%</td>
197
+ <td class="benchmark-value">73.3%</td>
198
+ <td class="benchmark-value">75.0%</td>
199
+ <td class="benchmark-value">93.0%</td>
200
+ </tr>
201
+ <tr>
202
+ <td class="benchmark-name">
203
+ <span class="benchmark-title">SWEBench Pro</span> <span class="benchmark-subtitle">(public)</span>
204
+ </td>
205
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">55.9%</span></td>
206
+ <td class="benchmark-value">50.9%</td>
207
+ <td class="benchmark-value">56.1%</td>
208
+ <td class="benchmark-value">56.2%</td>
209
+ <td class="benchmark-value">52.6%</td>
210
+ <td class="benchmark-value">46.4%</td>
211
+ <td class="benchmark-value">54.3%</td>
212
+ <td class="benchmark-value">39.5%</td>
213
+ <td class="benchmark-value">54.2%</td>
214
+ <td class="benchmark-value">62.7%</td>
215
+ </tr>
216
+ <tr>
217
+ <td class="benchmark-name">
218
+ <span class="benchmark-title">Terminal Bench 2.1</span> <span class="benchmark-subtitle">(best harness)</span>
219
+ </td>
220
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">64.7%</span></td>
221
+ <td class="benchmark-value">51.3%</td>
222
+ <td class="benchmark-value">63.7%</td>
223
+ <td class="benchmark-value">55.4%</td>
224
+ <td class="benchmark-value">61.8%</td>
225
+ <td class="benchmark-value">56.4%</td>
226
+ <td class="benchmark-value">63.8%</td>
227
+ <td class="benchmark-value">44.2%</td>
228
+ <td class="benchmark-value">54.0%</td>
229
+ <td class="benchmark-value">82.5%</td>
230
+ </tr>
231
+ <tr>
232
+ <td class="benchmark-name">
233
+ <span class="benchmark-title">SciCode</span>
234
+ </td>
235
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">48.7%</span></td>
236
+ <td class="benchmark-value">42.0%</td>
237
+ <td class="benchmark-value">43.1%</td>
238
+ <td class="benchmark-value">47.0%</td>
239
+ <td class="benchmark-value">44.9%</td>
240
+ <td class="benchmark-value">39.9%</td>
241
+ <td class="benchmark-value">46.1%</td>
242
+ <td class="benchmark-value">43.3%</td>
243
+ <td class="benchmark-value">40.9%</td>
244
+ <td class="benchmark-value">50.0%</td>
245
+ </tr>
246
+ </tbody>
247
+ <tbody class="benchmark-category-group">
248
+ <tr class="benchmark-category-row">
249
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Agentic (general)</span></th>
250
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
251
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
252
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
253
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
254
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
255
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
256
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
257
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
258
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
259
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
260
+ </tr>
261
+ <tr>
262
+ <td class="benchmark-name">
263
+ <span class="benchmark-title">GDPval-AA v2</span>
264
+ </td>
265
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">1269</span></td>
266
+ <td class="benchmark-value">962</td>
267
+ <td class="benchmark-value">1145</td>
268
+ <td class="benchmark-value">1159</td>
269
+ <td class="benchmark-value">1189</td>
270
+ <td class="benchmark-value">1164</td>
271
+ <td class="benchmark-value">1238</td>
272
+ <td class="benchmark-value">911</td>
273
+ <td class="benchmark-value">1139</td>
274
+ <td class="benchmark-value">1530</td>
275
+ </tr>
276
+ <tr>
277
+ <td class="benchmark-name">
278
+ <span class="benchmark-title">MCP Atlas</span> <span class="benchmark-subtitle">(public / all)</span>
279
+ </td>
280
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">79.6/79.2%</span></td>
281
+ <td class="benchmark-value">74.2%/–</td>
282
+ <td class="benchmark-value">–</td>
283
+ <td class="benchmark-value">49.4%/–</td>
284
+ <td class="benchmark-value">69.0%/–</td>
285
+ <td class="benchmark-value">47.4/44.7%</td>
286
+ <td class="benchmark-value">78.8/76.0%</td>
287
+ <td class="benchmark-value">41.2/40.2%</td>
288
+ <td class="benchmark-value">79.8/76.8%</td>
289
+ <td class="benchmark-value">77.0/75.0%</td>
290
+ </tr>
291
+ <tr>
292
+ <td class="benchmark-name">
293
+ <span class="benchmark-title">Tau 3 Banking</span>
294
+ </td>
295
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">15.5%</span></td>
296
+ <td class="benchmark-value">13.4%</td>
297
+ <td class="benchmark-value">6.6%</td>
298
+ <td class="benchmark-value">8.9%</td>
299
+ <td class="benchmark-value">22.9%</td>
300
+ <td class="benchmark-value">13.8%</td>
301
+ <td class="benchmark-value">23.7%</td>
302
+ <td class="benchmark-value">9.1%</td>
303
+ <td class="benchmark-value">16.5%</td>
304
+ <td class="benchmark-value">24.3%</td>
305
+ </tr>
306
+ <tr>
307
+ <td class="benchmark-name">
308
+ <span class="benchmark-title">BrowseComp</span> <span class="benchmark-subtitle">(with context management)</span>
309
+ </td>
310
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">77.4%</span></td>
311
+ <td class="benchmark-value">78.6%</td>
312
+ <td class="benchmark-value">–</td>
313
+ <td class="benchmark-value">76.3%</td>
314
+ <td class="benchmark-value">73.2%</td>
315
+ <td class="benchmark-value">63.0%</td>
316
+ <td class="benchmark-value">77.1%</td>
317
+ <td class="benchmark-value">–</td>
318
+ <td class="benchmark-value">–</td>
319
+ <td class="benchmark-value">84.0%</td>
320
+ </tr>
321
+ <tr>
322
+ <td class="benchmark-name">
323
+ <span class="benchmark-title">Toolathlon Verified</span>
324
+ </td>
325
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">54.4%</span></td>
326
+ <td class="benchmark-value">40.7%</td>
327
+ <td class="benchmark-value">49.1%</td>
328
+ <td class="benchmark-value">47.5%</td>
329
+ <td class="benchmark-value">50.9%</td>
330
+ <td class="benchmark-value">34.3%</td>
331
+ <td class="benchmark-value">45.5%</td>
332
+ <td class="benchmark-value">26.9%</td>
333
+ <td class="benchmark-value">57.1%</td>
334
+ <td class="benchmark-value">67.9%</td>
335
+ </tr>
336
+ <tr>
337
+ <td class="benchmark-name">
338
+ <span class="benchmark-title">AA-Briefcase</span>
339
+ </td>
340
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">917</span></td>
341
+ <td class="benchmark-value">–</td>
342
+ <td class="benchmark-value">–</td>
343
+ <td class="benchmark-value">–</td>
344
+ <td class="benchmark-value">833</td>
345
+ <td class="benchmark-value">870</td>
346
+ <td class="benchmark-value">839</td>
347
+ <td class="benchmark-value">612</td>
348
+ <td class="benchmark-value">–</td>
349
+ <td class="benchmark-value">–</td>
350
+ </tr>
351
+ </tbody>
352
+ <tbody class="benchmark-category-group">
353
+ <tr class="benchmark-category-row">
354
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Reasoning (general)</span></th>
355
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
356
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
357
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
358
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
359
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
360
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
361
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
362
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
363
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
364
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
365
+ </tr>
366
+ <tr>
367
+ <td class="benchmark-name">
368
+ <span class="benchmark-title">GPQA Diamond</span>
369
+ </td>
370
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">89.5%</span></td>
371
+ <td class="benchmark-value">89.3%</td>
372
+ <td class="benchmark-value">84.9%</td>
373
+ <td class="benchmark-value">87.4%</td>
374
+ <td class="benchmark-value">89.4%</td>
375
+ <td class="benchmark-value">86.7%</td>
376
+ <td class="benchmark-value">87.2%</td>
377
+ <td class="benchmark-value">67.2%</td>
378
+ <td class="benchmark-value">83.8%</td>
379
+ <td class="benchmark-value">89.5%</td>
380
+ </tr>
381
+ <tr>
382
+ <td class="benchmark-name">
383
+ <span class="benchmark-title">HLE</span> <span class="benchmark-subtitle">(text only)</span>
384
+ </td>
385
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">31.6%</span></td>
386
+ <td class="benchmark-value">27.3%</td>
387
+ <td class="benchmark-value">25.2%</td>
388
+ <td class="benchmark-value">28.1%</td>
389
+ <td class="benchmark-value">32.1%</td>
390
+ <td class="benchmark-value">26.6%</td>
391
+ <td class="benchmark-value">29.7%</td>
392
+ <td class="benchmark-value">9.7%</td>
393
+ <td class="benchmark-value">17.5%</td>
394
+ <td class="benchmark-value">35.6%</td>
395
+ </tr>
396
+ <tr>
397
+ <td class="benchmark-name">
398
+ <span class="benchmark-title">HLE</span> <span class="benchmark-subtitle">(with tools)</span>
399
+ </td>
400
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">47.8%</span></td>
401
+ <td class="benchmark-value">48.3%</td>
402
+ <td class="benchmark-value">40.0%</td>
403
+ <td class="benchmark-value">40.3%</td>
404
+ <td class="benchmark-value">45.1%</td>
405
+ <td class="benchmark-value">37.4%</td>
406
+ <td class="benchmark-value">46.0%</td>
407
+ <td class="benchmark-value">17.8%</td>
408
+ <td class="benchmark-value">42.5%</td>
409
+ <td class="benchmark-value">48.9%</td>
410
+ </tr>
411
+ <tr>
412
+ <td class="benchmark-name">
413
+ <span class="benchmark-title">AIME 2026</span>
414
+ </td>
415
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">95.5%</span></td>
416
+ <td class="benchmark-value">93.3%</td>
417
+ <td class="benchmark-value">93.6%</td>
418
+ <td class="benchmark-value">87.7%</td>
419
+ <td class="benchmark-value">95.8%</td>
420
+ <td class="benchmark-value">94.2%</td>
421
+ <td class="benchmark-value">97.1%</td>
422
+ <td class="benchmark-value">85.1%</td>
423
+ <td class="benchmark-value">82.2%</td>
424
+ <td class="benchmark-value">97.6%</td>
425
+ </tr>
426
+ <tr>
427
+ <td class="benchmark-name">
428
+ <span class="benchmark-title">HMMT Feb 2026</span>
429
+ </td>
430
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">90.2%</span></td>
431
+ <td class="benchmark-value">87.9%</td>
432
+ <td class="benchmark-value">82.6%</td>
433
+ <td class="benchmark-value">71.2%</td>
434
+ <td class="benchmark-value">93.9%</td>
435
+ <td class="benchmark-value">78.8%</td>
436
+ <td class="benchmark-value">86.3%</td>
437
+ <td class="benchmark-value">66.7%</td>
438
+ <td class="benchmark-value">63.6%</td>
439
+ <td class="benchmark-value">98.5%</td>
440
+ </tr>
441
+ <tr>
442
+ <td class="benchmark-name">
443
+ <span class="benchmark-title">CritPt</span>
444
+ </td>
445
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">8.3%</span></td>
446
+ <td class="benchmark-value">1.7%</td>
447
+ <td class="benchmark-value">3.7%</td>
448
+ <td class="benchmark-value">0.6%</td>
449
+ <td class="benchmark-value">7.1%</td>
450
+ <td class="benchmark-value">3.1%</td>
451
+ <td class="benchmark-value">5.4%</td>
452
+ <td class="benchmark-value">0.0%</td>
453
+ <td class="benchmark-value">0.0%</td>
454
+ <td class="benchmark-value">20.6%</td>
455
+ </tr>
456
+ </tbody>
457
+ <tbody class="benchmark-category-group">
458
+ <tr class="benchmark-category-row">
459
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Reasoning (abstract)</span></th>
460
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
461
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
462
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
463
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
464
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
465
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
466
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
467
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
468
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
469
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
470
+ </tr>
471
+ <tr>
472
+ <td class="benchmark-name">
473
+ <span class="benchmark-title">ARC-AGI-1</span>
474
+ </td>
475
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">84.0%</span></td>
476
+ <td class="benchmark-value">–</td>
477
+ <td class="benchmark-value">–</td>
478
+ <td class="benchmark-value">–</td>
479
+ <td class="benchmark-value">–</td>
480
+ <td class="benchmark-value">–</td>
481
+ <td class="benchmark-value">79.5%</td>
482
+ <td class="benchmark-value">47.7%</td>
483
+ <td class="benchmark-value">–</td>
484
+ <td class="benchmark-value">87.7%</td>
485
+ </tr>
486
+ <tr>
487
+ <td class="benchmark-name">
488
+ <span class="benchmark-title">ARC-AGI-2</span>
489
+ </td>
490
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">40.1%</span></td>
491
+ <td class="benchmark-value">–</td>
492
+ <td class="benchmark-value">–</td>
493
+ <td class="benchmark-value">–</td>
494
+ <td class="benchmark-value">–</td>
495
+ <td class="benchmark-value">–</td>
496
+ <td class="benchmark-value">36.5%</td>
497
+ <td class="benchmark-value">4.0%</td>
498
+ <td class="benchmark-value">–</td>
499
+ <td class="benchmark-value">47.6%</td>
500
+ </tr>
501
+ </tbody>
502
+ <tbody class="benchmark-category-group">
503
+ <tr class="benchmark-category-row">
504
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Factuality</span></th>
505
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
506
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
507
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
508
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
509
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
510
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
511
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
512
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
513
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
514
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
515
+ </tr>
516
+ <tr>
517
+ <td class="benchmark-name">
518
+ <span class="benchmark-title">SimpleQA Verified</span>
519
+ </td>
520
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">20.6%</span></td>
521
+ <td class="benchmark-value">26.0%</td>
522
+ <td class="benchmark-value">16.1%</td>
523
+ <td class="benchmark-value">13.5%</td>
524
+ <td class="benchmark-value">34.1%</td>
525
+ <td class="benchmark-value">32.4%</td>
526
+ <td class="benchmark-value">43.9%</td>
527
+ <td class="benchmark-value">5.9%</td>
528
+ <td class="benchmark-value">44.1%</td>
529
+ <td class="benchmark-value">41.7%</td>
530
+ </tr>
531
+ <tr>
532
+ <td class="benchmark-name">
533
+ <span class="benchmark-title">AA Omniscience</span> <span class="benchmark-subtitle">(index)</span>
534
+ </td>
535
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">-9.0</span></td>
536
+ <td class="benchmark-value">-29.8</td>
537
+ <td class="benchmark-value">-9.3</td>
538
+ <td class="benchmark-value">0.7</td>
539
+ <td class="benchmark-value">-22.9</td>
540
+ <td class="benchmark-value">-1.0</td>
541
+ <td class="benchmark-value">2.1</td>
542
+ <td class="benchmark-value">-4.2</td>
543
+ <td class="benchmark-value">6.9</td>
544
+ <td class="benchmark-value">-11.6</td>
545
+ </tr>
546
+ </tbody>
547
+ <tbody class="benchmark-category-group">
548
+ <tr class="benchmark-category-row">
549
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Chat</span></th>
550
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
551
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
552
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
553
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
554
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
555
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
556
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
557
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
558
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
559
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
560
+ </tr>
561
+ <tr>
562
+ <td class="benchmark-name">
563
+ <span class="benchmark-title">IFBench</span>
564
+ </td>
565
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">82.2%</span></td>
566
+ <td class="benchmark-value">78.8%</td>
567
+ <td class="benchmark-value">67.1%</td>
568
+ <td class="benchmark-value">75.7%</td>
569
+ <td class="benchmark-value">79.2%</td>
570
+ <td class="benchmark-value">81.4%</td>
571
+ <td class="benchmark-value">79.8%</td>
572
+ <td class="benchmark-value">54.3%</td>
573
+ <td class="benchmark-value">78.6%</td>
574
+ <td class="benchmark-value">67.3%</td>
575
+ </tr>
576
+ <tr>
577
+ <td class="benchmark-name">
578
+ <span class="benchmark-title">Global-MMLU-Lite</span>
579
+ </td>
580
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">86.7%</span></td>
581
+ <td class="benchmark-value">90.0%</td>
582
+ <td class="benchmark-value">83.5%</td>
583
+ <td class="benchmark-value">83.9%</td>
584
+ <td class="benchmark-value">88.4%</td>
585
+ <td class="benchmark-value">85.6%</td>
586
+ <td class="benchmark-value">88.7%</td>
587
+ <td class="benchmark-value">83.4%</td>
588
+ <td class="benchmark-value">89.4%</td>
589
+ <td class="benchmark-value">88.7%</td>
590
+ </tr>
591
+ </tbody>
592
+ <tbody class="benchmark-category-group">
593
+ <tr class="benchmark-category-row">
594
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Safety</span></th>
595
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
596
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
597
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
598
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
599
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
600
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
601
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
602
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
603
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
604
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
605
+ </tr>
606
+ <tr>
607
+ <td class="benchmark-name">
608
+ <span class="benchmark-title">StrongREJECT</span>
609
+ </td>
610
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">98.4%</span></td>
611
+ <td class="benchmark-value">99.4%</td>
612
+ <td class="benchmark-value">99.3%</td>
613
+ <td class="benchmark-value">99.4%</td>
614
+ <td class="benchmark-value">97.4%</td>
615
+ <td class="benchmark-value">98.7%</td>
616
+ <td class="benchmark-value">98.6%</td>
617
+ <td class="benchmark-value">98.6%</td>
618
+ <td class="benchmark-value">97.6%</td>
619
+ <td class="benchmark-value">98.7%</td>
620
+ </tr>
621
+ <tr>
622
+ <td class="benchmark-name">
623
+ <span class="benchmark-title">FORTRESS</span> <span class="benchmark-subtitle">(adversarial)</span>
624
+ </td>
625
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">71.6%</span></td>
626
+ <td class="benchmark-value">77.3%</td>
627
+ <td class="benchmark-value">64.8%</td>
628
+ <td class="benchmark-value">86.3%</td>
629
+ <td class="benchmark-value">32.0%</td>
630
+ <td class="benchmark-value">77.6%</td>
631
+ <td class="benchmark-value">78.0%</td>
632
+ <td class="benchmark-value">91.3%</td>
633
+ <td class="benchmark-value">70.7%</td>
634
+ <td class="benchmark-value">83.8%</td>
635
+ </tr>
636
+ <tr>
637
+ <td class="benchmark-name">
638
+ <span class="benchmark-title">FORTRESS</span> <span class="benchmark-subtitle">(benign)</span>
639
+ </td>
640
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">96.9%</span></td>
641
+ <td class="benchmark-value">95.4%</td>
642
+ <td class="benchmark-value">94.6%</td>
643
+ <td class="benchmark-value">90.1%</td>
644
+ <td class="benchmark-value">99.2%</td>
645
+ <td class="benchmark-value">90.6%</td>
646
+ <td class="benchmark-value">95.9%</td>
647
+ <td class="benchmark-value">94.1%</td>
648
+ <td class="benchmark-value">95.5%</td>
649
+ <td class="benchmark-value">97.8%</td>
650
+ </tr>
651
+ </tbody>
652
+ <tbody class="benchmark-category-group">
653
+ <tr class="benchmark-category-row">
654
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Vision</span></th>
655
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
656
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
657
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
658
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
659
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
660
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
661
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
662
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
663
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
664
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
665
+ </tr>
666
+ <tr>
667
+ <td class="benchmark-name">
668
+ <span class="benchmark-title">MMMU Pro</span> <span class="benchmark-subtitle">(Standard 10)</span>
669
+ </td>
670
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">74.0%</span></td>
671
+ <td class="benchmark-value">77.3%</td>
672
+ <td class="benchmark-value">75.4%</td>
673
+ <td class="benchmark-value">–</td>
674
+ <td class="benchmark-value">–</td>
675
+ <td class="benchmark-value">–</td>
676
+ <td class="benchmark-value">73.5%</td>
677
+ <td class="benchmark-value">58.6%</td>
678
+ <td class="benchmark-value">79.0%</td>
679
+ <td class="benchmark-value">78.6%</td>
680
+ </tr>
681
+ <tr>
682
+ <td class="benchmark-name">
683
+ <span class="benchmark-title">Charxiv RQ</span> <span class="benchmark-subtitle">(original / with python)</span>
684
+ </td>
685
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">77.4/81.3%</span></td>
686
+ <td class="benchmark-value">80.8%/–</td>
687
+ <td class="benchmark-value">81.0%/–</td>
688
+ <td class="benchmark-value">–</td>
689
+ <td class="benchmark-value">–</td>
690
+ <td class="benchmark-value">–</td>
691
+ <td class="benchmark-value">78.1/82.0%</td>
692
+ <td class="benchmark-value">57.4%/–</td>
693
+ <td class="benchmark-value">70.0%/–</td>
694
+ <td class="benchmark-value">81.4%/–</td>
695
+ </tr>
696
+ </tbody>
697
+ <tbody class="benchmark-category-group">
698
+ <tr class="benchmark-category-row">
699
+ <th class="benchmark-category-label" scope="rowgroup"><span class="benchmark-category-label-text">Audio</span></th>
700
+ <td class="benchmark-category-cell benchmark-value benchmark-instant-start" aria-hidden="true"></td>
701
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
702
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
703
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
704
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
705
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
706
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
707
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
708
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
709
+ <td class="benchmark-category-cell" aria-hidden="true"></td>
710
+ </tr>
711
+ <tr>
712
+ <td class="benchmark-name">
713
+ <span class="benchmark-title">Audio MC</span>
714
+ </td>
715
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">54.9%</span></td>
716
+ <td class="benchmark-value">–</td>
717
+ <td class="benchmark-value">30.4%</td>
718
+ <td class="benchmark-value">–</td>
719
+ <td class="benchmark-value">–</td>
720
+ <td class="benchmark-value">–</td>
721
+ <td class="benchmark-value">56.6%</td>
722
+ <td class="benchmark-value">–</td>
723
+ <td class="benchmark-value">33.6%</td>
724
+ <td class="benchmark-value">–</td>
725
+ </tr>
726
+ <tr>
727
+ <td class="benchmark-name">
728
+ <span class="benchmark-title">MMAU</span>
729
+ </td>
730
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">77.0%</span></td>
731
+ <td class="benchmark-value">–</td>
732
+ <td class="benchmark-value">73.6%</td>
733
+ <td class="benchmark-value">–</td>
734
+ <td class="benchmark-value">–</td>
735
+ <td class="benchmark-value">–</td>
736
+ <td class="benchmark-value">77.2%</td>
737
+ <td class="benchmark-value">–</td>
738
+ <td class="benchmark-value">75.2%</td>
739
+ <td class="benchmark-value">–</td>
740
+ </tr>
741
+ <tr>
742
+ <td class="benchmark-name">
743
+ <span class="benchmark-title">VoiceBench</span>
744
+ </td>
745
+ <td class="benchmark-value benchmark-instant-start"><span class="benchmark-lock-content">90.1%</span></td>
746
+ <td class="benchmark-value">–</td>
747
+ <td class="benchmark-value">86.4%</td>
748
+ <td class="benchmark-value">–</td>
749
+ <td class="benchmark-value">–</td>
750
+ <td class="benchmark-value">–</td>
751
+ <td class="benchmark-value">91.4%</td>
752
+ <td class="benchmark-value">–</td>
753
+ <td class="benchmark-value">85.9%</td>
754
+ <td class="benchmark-value">–</td>
755
+ </tr>
756
+ </tbody>
757
+ </table>
758
+
759
+ - <small>Inkling-Small against open-and closed-weights models across the full eval suite. Activated and total parameters are given for scale; a dash means the score was not available at the time of writing.</small>
760
+ - <small>SWEBench Verified: Inkling and Inkling-Small’s numbers are reported using a bash-only harness. We use self-reported numbers for external models.</small>
761
+ - <small>Terminal Bench 2.1: Inkling and Inkling-Small’s numbers are reported using an internal coding harness. A small number of solutions were found to be contaminated from web search and were assigned a score of 0. We use self-reported numbers for external models where available. Otherwise, we report performance using our internal harness.</small>
762
+ - <small>Audio MC: Other models were evaluated internally since they are not on the official leaderboard.</small>
763
+ - <small>VoiceBench: VoiceBench uses rule-based, hard-coded string matching for grading, making the evaluation sensitive to output-formatting differences. We therefore added a system message instructing models to follow the expected answer format.</small>
764
+ - <small>HLE with tools: We benchmarked Minimax M2.7, Claude 4.5 Haiku, Gemini 3.5 Flash-Lite, and GPT 5.6 Luna using our internal harness.</small>
765
+
766
+ ## 6. Safety
767
+
768
+ We conducted safety evaluations ahead of release, spanning both everyday human-AI interaction and dangerous-capability testing. Because Inkling-Small is multimodal, we paid attention to whether safety behavior held consistently across text, audio, and image inputs. We applied mitigations to reduce risks before release.
769
+
770
+ For everyday interaction, we evaluated sycophancy, harmful manipulation, and psychological-harm patterns like parasocial dependency and validation of delusional reasoning, including through multi-turn, open-ended external red-teaming designed to surface issues that only emerge over longer conversations. We also assessed whether the model refuses genuinely harmful requests without over-refusing benign ones. For CBRN and cyber, we assessed knowledge and procedural uplift through internal evaluations, external testing, and refusal-suppressed variants intended to estimate latent capability with safeguards removed. For loss of control, we evaluated agentic capability, strategic deception, and sabotage potential, benchmarked against public frontier models, and found the model materially below frontier capabilities.
771
+
772
+ Across all areas, we concluded that Inkling-Small did not present risk of material uplift beyond what's already available in the open-weight ecosystem.
773
+
774
+ The residual risks identified in our evaluations — specifically, Inkling-Small’s occasional tendency to comply with role-play and indirectly framed prompts concerning harmful topics — are consistent with what you would see from any open-weight model, and are best addressed with defense-in-depth rather than relying on the model's refusals alone. Common downstream moderation tools, such as Llama Guard, are compatible with Inkling-Small and can be layered around the model to catch jailbreak attempts, filter unsafe outputs, and enforce use-case-specific policies. We would encourage treating this kind of input/output classification as a part of your deployment stack, especially for consumer-facing or high-traffic applications where adversarial prompting is more likely.
775
+
776
+ ## 7. Bias, risks and limitations
777
+
778
+ Inkling-Small may exhibit general limitations common to foundation models, including hallucination (generating plausible but factually incorrect or unsupported content), occasional failures to follow instructions precisely, and degraded performance in long multi-turn conversations. As with other large-scale models trained on web-derived and synthetic data, Inkling-Small may reflect biases present in its training data, including demographic, cultural, or linguistic biases, and may perform unevenly across languages, dialects, or subject domains that were less represented during training.
779
+
780
+ Inkling-Small's knowledge is limited to information available as of its training cutoff, and it may not reflect events, developments, or changes that occurred afterward.
781
+
782
+ We recommend that downstream developers and deployers apply appropriate human oversight and review for outputs used in high-stakes or safety-critical contexts, rather than relying on Inkling-Small's outputs without verification.
783
+
784
+ - Conduct their own evaluation of Inkling-Small's performance, safety, and fairness for their specific use case, language, and population prior to deployment, particularly for applications involving vulnerable groups.
785
+ - Implement additional safeguards – such as content filtering, rate limiting, and monitoring – at the application layer, especially for open deployment contexts where Inkling-Small's built-in mitigations may not be sufficient on their own.
786
+ - Avoid deploying Inkling-Small in domains such as medical, legal, or safety-critical decision-making without additional fine-tuning, domain-specific validation, and human oversight
787
+
788
+ ## 8. Legal
789
+
790
+ [Training Data Documentation](https://thinkingmachines.ai/training-data-documentation/)
chat_template.jinja ADDED
@@ -0,0 +1,129 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- set effort_map = {"none": 0.0, "minimal": 0.1, "low": 0.2, "medium": 0.7, "high": 0.9, "max": 0.99} -%}
2
+ {%- set role_token = {"user": "<|message_user|>", "assistant": "<|message_model|>", "system": "<|message_system|>", "tool": "<|message_tool|>"} -%}
3
+
4
+ {%- macro emit_thinking_effort() -%}
5
+ {%- set eff = reasoning_effort if reasoning_effort is defined and reasoning_effort is not none else 0.9 -%}
6
+ {%- if eff is string -%}
7
+ {%- set key = eff | trim -%}
8
+ {%- if key not in effort_map -%}
9
+ {{- raise_exception("Unknown reasoning_effort: " ~ eff) -}}
10
+ {%- endif -%}
11
+ {%- set num = effort_map[key] -%}
12
+ {%- else -%}
13
+ {%- set num = eff | float -%}
14
+ {%- endif -%}
15
+ {%- if num < 0.0 or num > 0.99 -%}
16
+ {{- raise_exception("reasoning_effort must be in [0.0, 0.99]") -}}
17
+ {%- endif -%}
18
+ {{- "<|message_system|><|content_text|>Thinking effort level: " -}}
19
+ {%- if num == 0.0 -%}0{%- else -%}{{ num }}{%- endif -%}
20
+ {{- "<|end_message|>" -}}
21
+ {%- endmacro -%}
22
+
23
+ {%- if tools -%}
24
+ {%- set tool_state = namespace(specs=[]) -%}
25
+ {%- for tool in tools -%}
26
+ {%- set fn = tool.function if tool.function is defined else tool -%}
27
+ {%- set spec = {
28
+ "description": (fn.description if fn.description is defined and fn.description else ""),
29
+ "name": fn.name,
30
+ "parameters": (fn.parameters if fn.parameters is defined and fn.parameters else {}),
31
+ "type": (tool.type if tool.type is defined and tool.type else "function"),
32
+ } -%}
33
+ {%- set tool_state.specs = tool_state.specs + [spec] -%}
34
+ {%- endfor -%}
35
+ {{- "<|message_system|>tool_declare<|content_xml|>" -}}
36
+ {{- tool_state.specs | tojson(sort_keys=true, separators=(",", ":")) -}}
37
+ {{- "<|end_message|>" -}}
38
+ {%- endif -%}
39
+
40
+ {%- set state = namespace(effort_emitted=false) -%}
41
+ {%- for message in messages -%}
42
+ {%- if message.role not in role_token -%}
43
+ {{- raise_exception("Unknown message role: " ~ message.role) -}}
44
+ {%- endif -%}
45
+ {%- if not state.effort_emitted and message.role != "system" -%}
46
+ {{- emit_thinking_effort() -}}
47
+ {%- set state.effort_emitted = true -%}
48
+ {%- endif -%}
49
+
50
+ {%- set rtok = role_token[message.role] -%}
51
+
52
+ {%- if message.role == "tool" -%}
53
+ {%- set tool_name_state = namespace(name="") -%}
54
+ {%- if message.name is defined and message.name -%}
55
+ {%- set tool_name_state.name = message.name -%}
56
+ {%- elif message.tool_call_id is defined and message.tool_call_id -%}
57
+ {%- for prev in messages -%}
58
+ {%- if prev.role == "assistant" and prev.tool_calls -%}
59
+ {%- for tc in prev.tool_calls -%}
60
+ {%- if tc.id is defined and tc.id == message.tool_call_id and tc.function.name is defined -%}
61
+ {%- set tool_name_state.name = tc.function.name -%}
62
+ {%- endif -%}
63
+ {%- endfor -%}
64
+ {%- endif -%}
65
+ {%- endfor -%}
66
+ {%- endif -%}
67
+ {{- rtok -}}
68
+ {%- if tool_name_state.name -%}{{- tool_name_state.name -}}{%- endif -%}
69
+ {{- "<|content_text|>" -}}
70
+ {%- if message.content is string -%}{{- message.content -}}{%- endif -%}
71
+ {{- "<|end_message|>" -}}
72
+
73
+ {%- else -%}
74
+ {%- if message.role == "assistant" and message.reasoning_content is defined and message.reasoning_content -%}
75
+ {{- "<|message_model|><|content_thinking|>" ~ message.reasoning_content ~ "<|end_message|>" -}}
76
+ {%- endif -%}
77
+
78
+ {%- if message.content is string -%}
79
+ {{- rtok ~ "<|content_text|>" ~ message.content ~ "<|end_message|>" -}}
80
+ {%- elif message.content -%}
81
+ {%- for part in message.content -%}
82
+ {%- if part is string -%}
83
+ {{- rtok ~ "<|content_text|>" ~ part ~ "<|end_message|>" -}}
84
+ {%- elif part.type is not defined or part.type in ("text", "input_text") -%}
85
+ {%- set text_part = (part.text if part.text is defined and part.text is string else "") -%}
86
+ {{- rtok ~ "<|content_text|>" ~ text_part ~ "<|end_message|>" -}}
87
+ {%- elif part.type in ("image", "input_image", "image_url") -%}
88
+ {{- rtok ~ "<|content_image|><|unused_200054|><|end_message|>" -}}
89
+ {%- elif part.type in ("audio", "input_audio", "audio_url") -%}
90
+ {{- rtok ~ "<|content_audio_input|><|unused_200053|><|audio_end|><|end_message|>" -}}
91
+ {%- else -%}
92
+ {{- raise_exception("Unsupported content part type: " ~ part.type) -}}
93
+ {%- endif -%}
94
+ {%- endfor -%}
95
+ {%- endif -%}
96
+
97
+ {%- if message.role == "assistant" and message.tool_calls -%}
98
+ {%- for tc in message.tool_calls -%}
99
+ {%- set fn = tc.function -%}
100
+ {%- if fn.name is not defined or fn.name is not string -%}
101
+ {{- raise_exception("tool call function name must be a string") -}}
102
+ {%- endif -%}
103
+ {%- set args = fn.arguments if fn.arguments is defined and fn.arguments else {} -%}
104
+ {%- if args is string -%}
105
+ {{- raise_exception("tool call arguments must be a parsed object, not a JSON string; canonicalize upstream") -}}
106
+ {%- endif -%}
107
+ {%- if args is not mapping -%}
108
+ {{- raise_exception("tool call arguments must be an object") -}}
109
+ {%- endif -%}
110
+ {{- "<|message_model|>" ~ fn.name ~ "<|content_invoke_tool_json|>" -}}
111
+ {{- '{"name":' ~ (fn.name | tojson(sort_keys=true, separators=(",", ":"))) ~ ',"args":' -}}
112
+ {{- (args | tojson(sort_keys=true, separators=(",", ":"))) -}}
113
+ {{- "}<|end_message|>" -}}
114
+ {%- endfor -%}
115
+ {%- endif -%}
116
+
117
+ {%- if message.role == "assistant" -%}
118
+ {{- "<|content_model_end_sampling|>" -}}
119
+ {%- endif -%}
120
+ {%- endif -%}
121
+ {%- endfor -%}
122
+
123
+ {%- if not state.effort_emitted -%}
124
+ {{- emit_thinking_effort() -}}
125
+ {%- endif -%}
126
+
127
+ {%- if add_generation_prompt -%}
128
+ {{- "<|message_model|>" -}}
129
+ {%- endif -%}
config.json ADDED
@@ -0,0 +1,114 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "InklingForConditionalGeneration"
4
+ ],
5
+ "model_type": "inkling_mm_model",
6
+ "eos_token_id": 200006,
7
+ "text_config": {
8
+ "model_max_length": 1048576,
9
+ "torch_dtype": "bfloat16",
10
+ "hidden_size": 4096,
11
+ "num_hidden_layers": 42,
12
+ "vocab_size": 201024,
13
+ "num_attention_heads": 32,
14
+ "num_key_value_heads": 8,
15
+ "head_dim": 128,
16
+ "d_rel": 16,
17
+ "rel_extent": 1024,
18
+ "q_bias": false,
19
+ "o_bias": false,
20
+ "log_scaling_n_floor": 128000,
21
+ "log_scaling_alpha": 0.1,
22
+ "rms_norm_eps": 1e-06,
23
+ "use_embed_norm": true,
24
+ "local_layer_ids": [
25
+ 0,
26
+ 1,
27
+ 2,
28
+ 3,
29
+ 4,
30
+ 6,
31
+ 7,
32
+ 8,
33
+ 9,
34
+ 10,
35
+ 12,
36
+ 13,
37
+ 14,
38
+ 15,
39
+ 16,
40
+ 18,
41
+ 19,
42
+ 20,
43
+ 21,
44
+ 22,
45
+ 24,
46
+ 25,
47
+ 26,
48
+ 27,
49
+ 28,
50
+ 30,
51
+ 31,
52
+ 32,
53
+ 33,
54
+ 34,
55
+ 36,
56
+ 37,
57
+ 38,
58
+ 39,
59
+ 40
60
+ ],
61
+ "dense_mlp_idx": 2,
62
+ "use_sconv": true,
63
+ "sconv_kernel_size": 4,
64
+ "unpadded_vocab_size": 200058,
65
+ "logits_mup_width_multiplier": 16.0,
66
+ "final_logit_softcapping": null,
67
+ "swa_head_dim": 128,
68
+ "swa_num_attention_heads": 32,
69
+ "swa_num_key_value_heads": 8,
70
+ "sliding_window_size": 512,
71
+ "n_routed_experts": 256,
72
+ "num_experts_per_tok": 6,
73
+ "n_shared_experts": 2,
74
+ "shared_expert_sink": true,
75
+ "dense_intermediate_size": 16384,
76
+ "intermediate_size": 2048,
77
+ "route_scale": 8.0,
78
+ "use_gate_bias": true,
79
+ "gate_activation": "sigmoid",
80
+ "norm_after_topk": true,
81
+ "use_global_scale": true
82
+ },
83
+ "audio_config": {
84
+ "decoder_dmodel": 4096,
85
+ "n_mel_bins": 80,
86
+ "mel_vocab_size": 16,
87
+ "bias": false,
88
+ "dmel_min_value": -7.0,
89
+ "dmel_max_value": 2.0,
90
+ "use_audio_norm": true,
91
+ "audio_mode": "dmel"
92
+ },
93
+ "vision_config": {
94
+ "vision_encoder_type": "hmlp",
95
+ "decoder_dmodel": 4096,
96
+ "patch_size": 40,
97
+ "temporal_patch_size": 2,
98
+ "n_channels": 3,
99
+ "n_layers": 4,
100
+ "use_vision_norm": true
101
+ },
102
+ "mtp_config": {
103
+ "num_nextn_predict_layers": 8,
104
+ "chain_hidden_post_norm": false,
105
+ "local_layer_ids": [
106
+ 0,
107
+ 2,
108
+ 4,
109
+ 5,
110
+ 6,
111
+ 7
112
+ ]
113
+ }
114
+ }
hf_quant_config.json ADDED
@@ -0,0 +1,411 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "quantization": {
3
+ "quant_algo": "NVFP4",
4
+ "kv_cache_quant_algo": "none",
5
+ "exclude_modules": [
6
+ "model.audio.decoder",
7
+ "model.audio.encoder",
8
+ "model.audio.final_norm",
9
+ "model.llm.embed",
10
+ "model.llm.embed_norm",
11
+ "model.llm.norm",
12
+ "model.llm.unembed",
13
+ "model.visual.final_norm",
14
+ "model.visual.layers.linear_0",
15
+ "model.visual.layers.linear_1",
16
+ "model.visual.layers.linear_2",
17
+ "model.visual.layers.linear_3",
18
+ "model.visual.layers.norm_0",
19
+ "model.visual.layers.norm_1",
20
+ "model.visual.layers.norm_2",
21
+ "model.llm.layers.0.attn",
22
+ "model.llm.layers.0.attn_norm",
23
+ "model.llm.layers.0.attn_sconv",
24
+ "model.llm.layers.0.mlp.global_scale",
25
+ "model.llm.layers.0.mlp.w13_dn",
26
+ "model.llm.layers.0.mlp.w2_md",
27
+ "model.llm.layers.0.mlp_norm",
28
+ "model.llm.layers.0.mlp_sconv",
29
+ "model.llm.layers.1.attn",
30
+ "model.llm.layers.1.attn_norm",
31
+ "model.llm.layers.1.attn_sconv",
32
+ "model.llm.layers.1.mlp.global_scale",
33
+ "model.llm.layers.1.mlp.w13_dn",
34
+ "model.llm.layers.1.mlp.w2_md",
35
+ "model.llm.layers.1.mlp_norm",
36
+ "model.llm.layers.1.mlp_sconv",
37
+ "model.llm.layers.10.attn",
38
+ "model.llm.layers.10.attn_norm",
39
+ "model.llm.layers.10.attn_sconv",
40
+ "model.llm.layers.10.mlp.gate",
41
+ "model.llm.layers.10.mlp.shared_experts",
42
+ "model.llm.layers.10.mlp_norm",
43
+ "model.llm.layers.10.mlp_sconv",
44
+ "model.llm.layers.11.attn",
45
+ "model.llm.layers.11.attn_norm",
46
+ "model.llm.layers.11.attn_sconv",
47
+ "model.llm.layers.11.mlp.gate",
48
+ "model.llm.layers.11.mlp.shared_experts",
49
+ "model.llm.layers.11.mlp_norm",
50
+ "model.llm.layers.11.mlp_sconv",
51
+ "model.llm.layers.12.attn",
52
+ "model.llm.layers.12.attn_norm",
53
+ "model.llm.layers.12.attn_sconv",
54
+ "model.llm.layers.12.mlp.gate",
55
+ "model.llm.layers.12.mlp.shared_experts",
56
+ "model.llm.layers.12.mlp_norm",
57
+ "model.llm.layers.12.mlp_sconv",
58
+ "model.llm.layers.13.attn",
59
+ "model.llm.layers.13.attn_norm",
60
+ "model.llm.layers.13.attn_sconv",
61
+ "model.llm.layers.13.mlp.gate",
62
+ "model.llm.layers.13.mlp.shared_experts",
63
+ "model.llm.layers.13.mlp_norm",
64
+ "model.llm.layers.13.mlp_sconv",
65
+ "model.llm.layers.14.attn",
66
+ "model.llm.layers.14.attn_norm",
67
+ "model.llm.layers.14.attn_sconv",
68
+ "model.llm.layers.14.mlp.gate",
69
+ "model.llm.layers.14.mlp.shared_experts",
70
+ "model.llm.layers.14.mlp_norm",
71
+ "model.llm.layers.14.mlp_sconv",
72
+ "model.llm.layers.15.attn",
73
+ "model.llm.layers.15.attn_norm",
74
+ "model.llm.layers.15.attn_sconv",
75
+ "model.llm.layers.15.mlp.gate",
76
+ "model.llm.layers.15.mlp.shared_experts",
77
+ "model.llm.layers.15.mlp_norm",
78
+ "model.llm.layers.15.mlp_sconv",
79
+ "model.llm.layers.16.attn",
80
+ "model.llm.layers.16.attn_norm",
81
+ "model.llm.layers.16.attn_sconv",
82
+ "model.llm.layers.16.mlp.gate",
83
+ "model.llm.layers.16.mlp.shared_experts",
84
+ "model.llm.layers.16.mlp_norm",
85
+ "model.llm.layers.16.mlp_sconv",
86
+ "model.llm.layers.17.attn",
87
+ "model.llm.layers.17.attn_norm",
88
+ "model.llm.layers.17.attn_sconv",
89
+ "model.llm.layers.17.mlp.gate",
90
+ "model.llm.layers.17.mlp.shared_experts",
91
+ "model.llm.layers.17.mlp_norm",
92
+ "model.llm.layers.17.mlp_sconv",
93
+ "model.llm.layers.18.attn",
94
+ "model.llm.layers.18.attn_norm",
95
+ "model.llm.layers.18.attn_sconv",
96
+ "model.llm.layers.18.mlp.gate",
97
+ "model.llm.layers.18.mlp.shared_experts",
98
+ "model.llm.layers.18.mlp_norm",
99
+ "model.llm.layers.18.mlp_sconv",
100
+ "model.llm.layers.19.attn",
101
+ "model.llm.layers.19.attn_norm",
102
+ "model.llm.layers.19.attn_sconv",
103
+ "model.llm.layers.19.mlp.gate",
104
+ "model.llm.layers.19.mlp.shared_experts",
105
+ "model.llm.layers.19.mlp_norm",
106
+ "model.llm.layers.19.mlp_sconv",
107
+ "model.llm.layers.2.attn",
108
+ "model.llm.layers.2.attn_norm",
109
+ "model.llm.layers.2.attn_sconv",
110
+ "model.llm.layers.2.mlp.experts",
111
+ "model.llm.layers.2.mlp.gate",
112
+ "model.llm.layers.2.mlp.shared_experts",
113
+ "model.llm.layers.2.mlp_norm",
114
+ "model.llm.layers.2.mlp_sconv",
115
+ "model.llm.layers.20.attn",
116
+ "model.llm.layers.20.attn_norm",
117
+ "model.llm.layers.20.attn_sconv",
118
+ "model.llm.layers.20.mlp.gate",
119
+ "model.llm.layers.20.mlp.shared_experts",
120
+ "model.llm.layers.20.mlp_norm",
121
+ "model.llm.layers.20.mlp_sconv",
122
+ "model.llm.layers.21.attn",
123
+ "model.llm.layers.21.attn_norm",
124
+ "model.llm.layers.21.attn_sconv",
125
+ "model.llm.layers.21.mlp.gate",
126
+ "model.llm.layers.21.mlp.shared_experts",
127
+ "model.llm.layers.21.mlp_norm",
128
+ "model.llm.layers.21.mlp_sconv",
129
+ "model.llm.layers.22.attn",
130
+ "model.llm.layers.22.attn_norm",
131
+ "model.llm.layers.22.attn_sconv",
132
+ "model.llm.layers.22.mlp.gate",
133
+ "model.llm.layers.22.mlp.shared_experts",
134
+ "model.llm.layers.22.mlp_norm",
135
+ "model.llm.layers.22.mlp_sconv",
136
+ "model.llm.layers.23.attn",
137
+ "model.llm.layers.23.attn_norm",
138
+ "model.llm.layers.23.attn_sconv",
139
+ "model.llm.layers.23.mlp.gate",
140
+ "model.llm.layers.23.mlp.shared_experts",
141
+ "model.llm.layers.23.mlp_norm",
142
+ "model.llm.layers.23.mlp_sconv",
143
+ "model.llm.layers.24.attn",
144
+ "model.llm.layers.24.attn_norm",
145
+ "model.llm.layers.24.attn_sconv",
146
+ "model.llm.layers.24.mlp.gate",
147
+ "model.llm.layers.24.mlp.shared_experts",
148
+ "model.llm.layers.24.mlp_norm",
149
+ "model.llm.layers.24.mlp_sconv",
150
+ "model.llm.layers.25.attn",
151
+ "model.llm.layers.25.attn_norm",
152
+ "model.llm.layers.25.attn_sconv",
153
+ "model.llm.layers.25.mlp.gate",
154
+ "model.llm.layers.25.mlp.shared_experts",
155
+ "model.llm.layers.25.mlp_norm",
156
+ "model.llm.layers.25.mlp_sconv",
157
+ "model.llm.layers.26.attn",
158
+ "model.llm.layers.26.attn_norm",
159
+ "model.llm.layers.26.attn_sconv",
160
+ "model.llm.layers.26.mlp.gate",
161
+ "model.llm.layers.26.mlp.shared_experts",
162
+ "model.llm.layers.26.mlp_norm",
163
+ "model.llm.layers.26.mlp_sconv",
164
+ "model.llm.layers.27.attn",
165
+ "model.llm.layers.27.attn_norm",
166
+ "model.llm.layers.27.attn_sconv",
167
+ "model.llm.layers.27.mlp.gate",
168
+ "model.llm.layers.27.mlp.shared_experts",
169
+ "model.llm.layers.27.mlp_norm",
170
+ "model.llm.layers.27.mlp_sconv",
171
+ "model.llm.layers.28.attn",
172
+ "model.llm.layers.28.attn_norm",
173
+ "model.llm.layers.28.attn_sconv",
174
+ "model.llm.layers.28.mlp.gate",
175
+ "model.llm.layers.28.mlp.shared_experts",
176
+ "model.llm.layers.28.mlp_norm",
177
+ "model.llm.layers.28.mlp_sconv",
178
+ "model.llm.layers.29.attn",
179
+ "model.llm.layers.29.attn_norm",
180
+ "model.llm.layers.29.attn_sconv",
181
+ "model.llm.layers.29.mlp.gate",
182
+ "model.llm.layers.29.mlp.shared_experts",
183
+ "model.llm.layers.29.mlp_norm",
184
+ "model.llm.layers.29.mlp_sconv",
185
+ "model.llm.layers.3.attn",
186
+ "model.llm.layers.3.attn_norm",
187
+ "model.llm.layers.3.attn_sconv",
188
+ "model.llm.layers.3.mlp.gate",
189
+ "model.llm.layers.3.mlp.shared_experts",
190
+ "model.llm.layers.3.mlp_norm",
191
+ "model.llm.layers.3.mlp_sconv",
192
+ "model.llm.layers.30.attn",
193
+ "model.llm.layers.30.attn_norm",
194
+ "model.llm.layers.30.attn_sconv",
195
+ "model.llm.layers.30.mlp.gate",
196
+ "model.llm.layers.30.mlp.shared_experts",
197
+ "model.llm.layers.30.mlp_norm",
198
+ "model.llm.layers.30.mlp_sconv",
199
+ "model.llm.layers.31.attn",
200
+ "model.llm.layers.31.attn_norm",
201
+ "model.llm.layers.31.attn_sconv",
202
+ "model.llm.layers.31.mlp.gate",
203
+ "model.llm.layers.31.mlp.shared_experts",
204
+ "model.llm.layers.31.mlp_norm",
205
+ "model.llm.layers.31.mlp_sconv",
206
+ "model.llm.layers.32.attn",
207
+ "model.llm.layers.32.attn_norm",
208
+ "model.llm.layers.32.attn_sconv",
209
+ "model.llm.layers.32.mlp.gate",
210
+ "model.llm.layers.32.mlp.shared_experts",
211
+ "model.llm.layers.32.mlp_norm",
212
+ "model.llm.layers.32.mlp_sconv",
213
+ "model.llm.layers.33.attn",
214
+ "model.llm.layers.33.attn_norm",
215
+ "model.llm.layers.33.attn_sconv",
216
+ "model.llm.layers.33.mlp.gate",
217
+ "model.llm.layers.33.mlp.shared_experts",
218
+ "model.llm.layers.33.mlp_norm",
219
+ "model.llm.layers.33.mlp_sconv",
220
+ "model.llm.layers.34.attn",
221
+ "model.llm.layers.34.attn_norm",
222
+ "model.llm.layers.34.attn_sconv",
223
+ "model.llm.layers.34.mlp.gate",
224
+ "model.llm.layers.34.mlp.shared_experts",
225
+ "model.llm.layers.34.mlp_norm",
226
+ "model.llm.layers.34.mlp_sconv",
227
+ "model.llm.layers.35.attn",
228
+ "model.llm.layers.35.attn_norm",
229
+ "model.llm.layers.35.attn_sconv",
230
+ "model.llm.layers.35.mlp.gate",
231
+ "model.llm.layers.35.mlp.shared_experts",
232
+ "model.llm.layers.35.mlp_norm",
233
+ "model.llm.layers.35.mlp_sconv",
234
+ "model.llm.layers.36.attn",
235
+ "model.llm.layers.36.attn_norm",
236
+ "model.llm.layers.36.attn_sconv",
237
+ "model.llm.layers.36.mlp.gate",
238
+ "model.llm.layers.36.mlp.shared_experts",
239
+ "model.llm.layers.36.mlp_norm",
240
+ "model.llm.layers.36.mlp_sconv",
241
+ "model.llm.layers.37.attn",
242
+ "model.llm.layers.37.attn_norm",
243
+ "model.llm.layers.37.attn_sconv",
244
+ "model.llm.layers.37.mlp.gate",
245
+ "model.llm.layers.37.mlp.shared_experts",
246
+ "model.llm.layers.37.mlp_norm",
247
+ "model.llm.layers.37.mlp_sconv",
248
+ "model.llm.layers.38.attn",
249
+ "model.llm.layers.38.attn_norm",
250
+ "model.llm.layers.38.attn_sconv",
251
+ "model.llm.layers.38.mlp.gate",
252
+ "model.llm.layers.38.mlp.shared_experts",
253
+ "model.llm.layers.38.mlp_norm",
254
+ "model.llm.layers.38.mlp_sconv",
255
+ "model.llm.layers.39.attn",
256
+ "model.llm.layers.39.attn_norm",
257
+ "model.llm.layers.39.attn_sconv",
258
+ "model.llm.layers.39.mlp.gate",
259
+ "model.llm.layers.39.mlp.shared_experts",
260
+ "model.llm.layers.39.mlp_norm",
261
+ "model.llm.layers.39.mlp_sconv",
262
+ "model.llm.layers.4.attn",
263
+ "model.llm.layers.4.attn_norm",
264
+ "model.llm.layers.4.attn_sconv",
265
+ "model.llm.layers.4.mlp.gate",
266
+ "model.llm.layers.4.mlp.shared_experts",
267
+ "model.llm.layers.4.mlp_norm",
268
+ "model.llm.layers.4.mlp_sconv",
269
+ "model.llm.layers.40.attn",
270
+ "model.llm.layers.40.attn_norm",
271
+ "model.llm.layers.40.attn_sconv",
272
+ "model.llm.layers.40.mlp.gate",
273
+ "model.llm.layers.40.mlp.shared_experts",
274
+ "model.llm.layers.40.mlp_norm",
275
+ "model.llm.layers.40.mlp_sconv",
276
+ "model.llm.layers.41.attn",
277
+ "model.llm.layers.41.attn_norm",
278
+ "model.llm.layers.41.attn_sconv",
279
+ "model.llm.layers.41.mlp.gate",
280
+ "model.llm.layers.41.mlp.shared_experts",
281
+ "model.llm.layers.41.mlp_norm",
282
+ "model.llm.layers.41.mlp_sconv",
283
+ "model.llm.layers.5.attn",
284
+ "model.llm.layers.5.attn_norm",
285
+ "model.llm.layers.5.attn_sconv",
286
+ "model.llm.layers.5.mlp.gate",
287
+ "model.llm.layers.5.mlp.shared_experts",
288
+ "model.llm.layers.5.mlp_norm",
289
+ "model.llm.layers.5.mlp_sconv",
290
+ "model.llm.layers.6.attn",
291
+ "model.llm.layers.6.attn_norm",
292
+ "model.llm.layers.6.attn_sconv",
293
+ "model.llm.layers.6.mlp.gate",
294
+ "model.llm.layers.6.mlp.shared_experts",
295
+ "model.llm.layers.6.mlp_norm",
296
+ "model.llm.layers.6.mlp_sconv",
297
+ "model.llm.layers.7.attn",
298
+ "model.llm.layers.7.attn_norm",
299
+ "model.llm.layers.7.attn_sconv",
300
+ "model.llm.layers.7.mlp.gate",
301
+ "model.llm.layers.7.mlp.shared_experts",
302
+ "model.llm.layers.7.mlp_norm",
303
+ "model.llm.layers.7.mlp_sconv",
304
+ "model.llm.layers.8.attn",
305
+ "model.llm.layers.8.attn_norm",
306
+ "model.llm.layers.8.attn_sconv",
307
+ "model.llm.layers.8.mlp.gate",
308
+ "model.llm.layers.8.mlp.shared_experts",
309
+ "model.llm.layers.8.mlp_norm",
310
+ "model.llm.layers.8.mlp_sconv",
311
+ "model.llm.layers.9.attn",
312
+ "model.llm.layers.9.attn_norm",
313
+ "model.llm.layers.9.attn_sconv",
314
+ "model.llm.layers.9.mlp.gate",
315
+ "model.llm.layers.9.mlp.shared_experts",
316
+ "model.llm.layers.9.mlp_norm",
317
+ "model.llm.layers.9.mlp_sconv"
318
+ ],
319
+ "modelopt_quant_config": {
320
+ "quant_cfg": {
321
+ "*weight_quantizer": {
322
+ "num_bits": [
323
+ 2,
324
+ 1
325
+ ],
326
+ "block_sizes": {
327
+ "-1": 16,
328
+ "type": "dynamic",
329
+ "scale_bits": [
330
+ 4,
331
+ 3
332
+ ]
333
+ },
334
+ "axis": null,
335
+ "enable": true
336
+ },
337
+ "*input_quantizer": {
338
+ "num_bits": [
339
+ 2,
340
+ 1
341
+ ],
342
+ "block_sizes": {
343
+ "-1": 16,
344
+ "type": "dynamic",
345
+ "scale_bits": [
346
+ 4,
347
+ 3
348
+ ]
349
+ },
350
+ "axis": null,
351
+ "enable": true
352
+ },
353
+ "nn.BatchNorm1d": {
354
+ "*": {
355
+ "enable": false
356
+ }
357
+ },
358
+ "nn.BatchNorm2d": {
359
+ "*": {
360
+ "enable": false
361
+ }
362
+ },
363
+ "nn.BatchNorm3d": {
364
+ "*": {
365
+ "enable": false
366
+ }
367
+ },
368
+ "nn.LeakyReLU": {
369
+ "*": {
370
+ "enable": false
371
+ }
372
+ },
373
+ "*lm_head*": {
374
+ "enable": false
375
+ },
376
+ "*proj_out.*": {
377
+ "enable": false
378
+ },
379
+ "*block_sparse_moe.gate*": {
380
+ "enable": false
381
+ },
382
+ "*router*": {
383
+ "enable": false
384
+ },
385
+ "*mlp.gate.*": {
386
+ "enable": false
387
+ },
388
+ "*mlp.shared_expert_gate.*": {
389
+ "enable": false
390
+ },
391
+ "*linear_attn.conv1d*": {
392
+ "enable": false
393
+ },
394
+ "*mixer.conv1d*": {
395
+ "enable": false
396
+ },
397
+ "*output_layer*": {
398
+ "enable": false
399
+ },
400
+ "output.*": {
401
+ "enable": false
402
+ },
403
+ "default": {
404
+ "enable": false
405
+ }
406
+ },
407
+ "algorithm": "max"
408
+ },
409
+ "group_size": 16
410
+ }
411
+ }
model-00001-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0536fc0774c061f006fbc0f8062300b80f7fa3c6395753f9558a53231ccdfdaf
3
+ size 19797851172
model-00002-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:94046cd4c60e9f8b84e646af29b48d972b16328ef16c68e1d224e7cc1c108f69
3
+ size 18832947622
model-00003-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5b011d77adab48cfedab8d82da3ee45e9658a298b5a2efe79386c9f80be83c90
3
+ size 19460482192
model-00004-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f84efa4eab97da5d95e42813cd40a682439cfbd6cbdc9e644cc2aed650034b66
3
+ size 19886208500
model-00005-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:04911920af137efb2bac5a8a3c65b1cd727b1c8f0a6ddcc691534c52e5cb3dce
3
+ size 19970085622
model-00006-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f2e9eb03063a3cdbc9361fd7f203d213476d7ca3295fe45757e2fe1f26513a8
3
+ size 18910731870
model-00007-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:728792823d6ba5b214503a3d75d39bbd239ba129a40b34ddb67f6d93e03fc0df
3
+ size 15485561834
model-00008-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc821c24f1af8342e02a456f06315c8c6c1e374d2060e258864117a660fbf37e
3
+ size 19791451186
model-00009-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c91d957b4a0cf277822f4174a5bc6071eec51b0c7243f53069c6f3e3bb8a93fb
3
+ size 14134068242
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
mtp.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d286dd21cb982a0052d24ee0077ec6fc38f5a766dc6953e6c4b85c6473bfa7b3
3
+ size 4463845392
processor_config.json ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_token": "<|unused_200053|>",
3
+ "audio_bos_token": "<|content_audio_input|>",
4
+ "dmel_max_value": 2.0,
5
+ "dmel_min_value": -7.0,
6
+ "feature_extractor": {
7
+ "audio_token_duration_s": 0.05,
8
+ "feature_extractor_type": "InklingFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 800,
11
+ "n_fft": 1600,
12
+ "padding_side": "right",
13
+ "padding_value": 0.0,
14
+ "return_attention_mask": true,
15
+ "sampling_rate": 16000,
16
+ "window_size": 1600,
17
+ "window_size_multiplier": 2.0
18
+ },
19
+ "image_processor": {
20
+ "do_convert_rgb": true,
21
+ "do_normalize": true,
22
+ "do_rescale": true,
23
+ "do_resize": true,
24
+ "image_mean": [
25
+ 0.48145466,
26
+ 0.4578275,
27
+ 0.40821073
28
+ ],
29
+ "image_processor_type": "InklingImageProcessor",
30
+ "image_std": [
31
+ 0.26862954,
32
+ 0.26130258,
33
+ 0.27577711
34
+ ],
35
+ "resample": 3,
36
+ "rescale_factor": 0.00392156862745098,
37
+ "size": {
38
+ "height": 40,
39
+ "width": 40
40
+ }
41
+ },
42
+ "image_token": "<|unused_200054|>",
43
+ "image_bos_token": "<|content_image|>",
44
+ "num_dmel_bins": 16,
45
+ "processor_class": "InklingProcessor"
46
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|endoftext|>",
4
+ "<|message_user|>",
5
+ "<|message_model|>",
6
+ "<|message_system|>",
7
+ "<|message_tool|>",
8
+ "<|content_text|>",
9
+ "<|content_image|>",
10
+ "<|content_model_end_sampling|>",
11
+ "<|content_thinking|>",
12
+ "<|end_message|>",
13
+ "<|content_audio_input|>",
14
+ "<|content_tool_error|>",
15
+ "<|audio|>",
16
+ "<|content_xml|>",
17
+ "<|begin_of_text|>",
18
+ "<|audio_end|>",
19
+ "<|content_invoke_tool_json|>",
20
+ "<|content_invoke_tool_text|>"
21
+ ]
22
+ }
tiktoken/tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bc253fd2b702f7a6da7105eaa8f3463b2f1247e83614f23e5323b921088bed2a
3
+ size 3615874
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9fb6333a7db8fe5da90728e741e4a3ee4ac2ae12c5dd4958cc6f31688787d3c2
3
+ size 27875797
tokenizer_config.json ADDED
@@ -0,0 +1,508 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "199998": {
4
+ "content": "<|unused|>",
5
+ "single_word": false,
6
+ "lstrip": false,
7
+ "rstrip": false,
8
+ "normalized": false,
9
+ "special": true
10
+ },
11
+ "199999": {
12
+ "content": "<|endoftext|>",
13
+ "single_word": false,
14
+ "lstrip": false,
15
+ "rstrip": false,
16
+ "normalized": false,
17
+ "special": true
18
+ },
19
+ "200000": {
20
+ "content": "<|message_user|>",
21
+ "single_word": false,
22
+ "lstrip": false,
23
+ "rstrip": false,
24
+ "normalized": false,
25
+ "special": true
26
+ },
27
+ "200001": {
28
+ "content": "<|message_model|>",
29
+ "single_word": false,
30
+ "lstrip": false,
31
+ "rstrip": false,
32
+ "normalized": false,
33
+ "special": true
34
+ },
35
+ "200002": {
36
+ "content": "<|message_system|>",
37
+ "single_word": false,
38
+ "lstrip": false,
39
+ "rstrip": false,
40
+ "normalized": false,
41
+ "special": true
42
+ },
43
+ "200003": {
44
+ "content": "<|message_tool|>",
45
+ "single_word": false,
46
+ "lstrip": false,
47
+ "rstrip": false,
48
+ "normalized": false,
49
+ "special": true
50
+ },
51
+ "200004": {
52
+ "content": "<|content_text|>",
53
+ "single_word": false,
54
+ "lstrip": false,
55
+ "rstrip": false,
56
+ "normalized": false,
57
+ "special": true
58
+ },
59
+ "200005": {
60
+ "content": "<|content_image|>",
61
+ "single_word": false,
62
+ "lstrip": false,
63
+ "rstrip": false,
64
+ "normalized": false,
65
+ "special": true
66
+ },
67
+ "200006": {
68
+ "content": "<|content_model_end_sampling|>",
69
+ "single_word": false,
70
+ "lstrip": false,
71
+ "rstrip": false,
72
+ "normalized": false,
73
+ "special": true
74
+ },
75
+ "200007": {
76
+ "content": "<|unused_200007|>",
77
+ "single_word": false,
78
+ "lstrip": false,
79
+ "rstrip": false,
80
+ "normalized": false,
81
+ "special": true
82
+ },
83
+ "200008": {
84
+ "content": "<|content_thinking|>",
85
+ "single_word": false,
86
+ "lstrip": false,
87
+ "rstrip": false,
88
+ "normalized": false,
89
+ "special": true
90
+ },
91
+ "200009": {
92
+ "content": "<|unused_200009|>",
93
+ "single_word": false,
94
+ "lstrip": false,
95
+ "rstrip": false,
96
+ "normalized": false,
97
+ "special": true
98
+ },
99
+ "200010": {
100
+ "content": "<|end_message|>",
101
+ "single_word": false,
102
+ "lstrip": false,
103
+ "rstrip": false,
104
+ "normalized": false,
105
+ "special": true
106
+ },
107
+ "200011": {
108
+ "content": "<|unused_200011|>",
109
+ "single_word": false,
110
+ "lstrip": false,
111
+ "rstrip": false,
112
+ "normalized": false,
113
+ "special": true
114
+ },
115
+ "200012": {
116
+ "content": "<|unused_200012|>",
117
+ "single_word": false,
118
+ "lstrip": false,
119
+ "rstrip": false,
120
+ "normalized": false,
121
+ "special": true
122
+ },
123
+ "200013": {
124
+ "content": "<|unused_200013|>",
125
+ "single_word": false,
126
+ "lstrip": false,
127
+ "rstrip": false,
128
+ "normalized": false,
129
+ "special": true
130
+ },
131
+ "200014": {
132
+ "content": "<|unused_200014|>",
133
+ "single_word": false,
134
+ "lstrip": false,
135
+ "rstrip": false,
136
+ "normalized": false,
137
+ "special": true
138
+ },
139
+ "200015": {
140
+ "content": "<|unused_200015|>",
141
+ "single_word": false,
142
+ "lstrip": false,
143
+ "rstrip": false,
144
+ "normalized": false,
145
+ "special": true
146
+ },
147
+ "200016": {
148
+ "content": "<|unused_200016|>",
149
+ "single_word": false,
150
+ "lstrip": false,
151
+ "rstrip": false,
152
+ "normalized": false,
153
+ "special": true
154
+ },
155
+ "200017": {
156
+ "content": "<|unused_200017|>",
157
+ "single_word": false,
158
+ "lstrip": false,
159
+ "rstrip": false,
160
+ "normalized": false,
161
+ "special": true
162
+ },
163
+ "200018": {
164
+ "content": "<|unused_200018|>",
165
+ "single_word": false,
166
+ "lstrip": false,
167
+ "rstrip": false,
168
+ "normalized": false,
169
+ "special": true
170
+ },
171
+ "200019": {
172
+ "content": "<|unused_200019|>",
173
+ "single_word": false,
174
+ "lstrip": false,
175
+ "rstrip": false,
176
+ "normalized": false,
177
+ "special": true
178
+ },
179
+ "200020": {
180
+ "content": "<|content_audio_input|>",
181
+ "single_word": false,
182
+ "lstrip": false,
183
+ "rstrip": false,
184
+ "normalized": false,
185
+ "special": true
186
+ },
187
+ "200021": {
188
+ "content": "<|unused_200021|>",
189
+ "single_word": false,
190
+ "lstrip": false,
191
+ "rstrip": false,
192
+ "normalized": false,
193
+ "special": true
194
+ },
195
+ "200022": {
196
+ "content": "<|content_tool_error|>",
197
+ "single_word": false,
198
+ "lstrip": false,
199
+ "rstrip": false,
200
+ "normalized": false,
201
+ "special": true
202
+ },
203
+ "200023": {
204
+ "content": "<|audio|>",
205
+ "single_word": false,
206
+ "lstrip": false,
207
+ "rstrip": false,
208
+ "normalized": false,
209
+ "special": true
210
+ },
211
+ "200024": {
212
+ "content": "<|content_xml|>",
213
+ "single_word": false,
214
+ "lstrip": false,
215
+ "rstrip": false,
216
+ "normalized": false,
217
+ "special": true
218
+ },
219
+ "200025": {
220
+ "content": "<|unused_200025|>",
221
+ "single_word": false,
222
+ "lstrip": false,
223
+ "rstrip": false,
224
+ "normalized": false,
225
+ "special": true
226
+ },
227
+ "200026": {
228
+ "content": "<|unused_200026|>",
229
+ "single_word": false,
230
+ "lstrip": false,
231
+ "rstrip": false,
232
+ "normalized": false,
233
+ "special": true
234
+ },
235
+ "200027": {
236
+ "content": "<|unused_200027|>",
237
+ "single_word": false,
238
+ "lstrip": false,
239
+ "rstrip": false,
240
+ "normalized": false,
241
+ "special": true
242
+ },
243
+ "200028": {
244
+ "content": "<|begin_of_text|>",
245
+ "single_word": false,
246
+ "lstrip": false,
247
+ "rstrip": false,
248
+ "normalized": false,
249
+ "special": true
250
+ },
251
+ "200029": {
252
+ "content": "<|unused_200029|>",
253
+ "single_word": false,
254
+ "lstrip": false,
255
+ "rstrip": false,
256
+ "normalized": false,
257
+ "special": true
258
+ },
259
+ "200030": {
260
+ "content": "<|unused_200030|>",
261
+ "single_word": false,
262
+ "lstrip": false,
263
+ "rstrip": false,
264
+ "normalized": false,
265
+ "special": true
266
+ },
267
+ "200031": {
268
+ "content": "<|unused_200031|>",
269
+ "single_word": false,
270
+ "lstrip": false,
271
+ "rstrip": false,
272
+ "normalized": false,
273
+ "special": true
274
+ },
275
+ "200032": {
276
+ "content": "<|unused_200032|>",
277
+ "single_word": false,
278
+ "lstrip": false,
279
+ "rstrip": false,
280
+ "normalized": false,
281
+ "special": true
282
+ },
283
+ "200033": {
284
+ "content": "<|unused_200033|>",
285
+ "single_word": false,
286
+ "lstrip": false,
287
+ "rstrip": false,
288
+ "normalized": false,
289
+ "special": true
290
+ },
291
+ "200034": {
292
+ "content": "<|unused_200034|>",
293
+ "single_word": false,
294
+ "lstrip": false,
295
+ "rstrip": false,
296
+ "normalized": false,
297
+ "special": true
298
+ },
299
+ "200035": {
300
+ "content": "<|unused_200035|>",
301
+ "single_word": false,
302
+ "lstrip": false,
303
+ "rstrip": false,
304
+ "normalized": false,
305
+ "special": true
306
+ },
307
+ "200036": {
308
+ "content": "<|unused_200036|>",
309
+ "single_word": false,
310
+ "lstrip": false,
311
+ "rstrip": false,
312
+ "normalized": false,
313
+ "special": true
314
+ },
315
+ "200037": {
316
+ "content": "<|unused_200037|>",
317
+ "single_word": false,
318
+ "lstrip": false,
319
+ "rstrip": false,
320
+ "normalized": false,
321
+ "special": true
322
+ },
323
+ "200038": {
324
+ "content": "<|unused_200038|>",
325
+ "single_word": false,
326
+ "lstrip": false,
327
+ "rstrip": false,
328
+ "normalized": false,
329
+ "special": true
330
+ },
331
+ "200039": {
332
+ "content": "<|unused_200039|>",
333
+ "single_word": false,
334
+ "lstrip": false,
335
+ "rstrip": false,
336
+ "normalized": false,
337
+ "special": true
338
+ },
339
+ "200040": {
340
+ "content": "<|unused_200040|>",
341
+ "single_word": false,
342
+ "lstrip": false,
343
+ "rstrip": false,
344
+ "normalized": false,
345
+ "special": true
346
+ },
347
+ "200041": {
348
+ "content": "<|unused_200041|>",
349
+ "single_word": false,
350
+ "lstrip": false,
351
+ "rstrip": false,
352
+ "normalized": false,
353
+ "special": true
354
+ },
355
+ "200042": {
356
+ "content": "<|unused_200042|>",
357
+ "single_word": false,
358
+ "lstrip": false,
359
+ "rstrip": false,
360
+ "normalized": false,
361
+ "special": true
362
+ },
363
+ "200043": {
364
+ "content": "<|audio_end|>",
365
+ "single_word": false,
366
+ "lstrip": false,
367
+ "rstrip": false,
368
+ "normalized": false,
369
+ "special": true
370
+ },
371
+ "200044": {
372
+ "content": "<|unused_200044|>",
373
+ "single_word": false,
374
+ "lstrip": false,
375
+ "rstrip": false,
376
+ "normalized": false,
377
+ "special": true
378
+ },
379
+ "200045": {
380
+ "content": "<|unused_200045|>",
381
+ "single_word": false,
382
+ "lstrip": false,
383
+ "rstrip": false,
384
+ "normalized": false,
385
+ "special": true
386
+ },
387
+ "200046": {
388
+ "content": "<|unused_200046|>",
389
+ "single_word": false,
390
+ "lstrip": false,
391
+ "rstrip": false,
392
+ "normalized": false,
393
+ "special": true
394
+ },
395
+ "200047": {
396
+ "content": "<|unused_200047|>",
397
+ "single_word": false,
398
+ "lstrip": false,
399
+ "rstrip": false,
400
+ "normalized": false,
401
+ "special": true
402
+ },
403
+ "200048": {
404
+ "content": "<|unused_200048|>",
405
+ "single_word": false,
406
+ "lstrip": false,
407
+ "rstrip": false,
408
+ "normalized": false,
409
+ "special": true
410
+ },
411
+ "200049": {
412
+ "content": "<|content_invoke_tool_json|>",
413
+ "single_word": false,
414
+ "lstrip": false,
415
+ "rstrip": false,
416
+ "normalized": false,
417
+ "special": true
418
+ },
419
+ "200050": {
420
+ "content": "<|unused_200050|>",
421
+ "single_word": false,
422
+ "lstrip": false,
423
+ "rstrip": false,
424
+ "normalized": false,
425
+ "special": true
426
+ },
427
+ "200051": {
428
+ "content": "<|unused_200051|>",
429
+ "single_word": false,
430
+ "lstrip": false,
431
+ "rstrip": false,
432
+ "normalized": false,
433
+ "special": true
434
+ },
435
+ "200052": {
436
+ "content": "<|unused_200052|>",
437
+ "single_word": false,
438
+ "lstrip": false,
439
+ "rstrip": false,
440
+ "normalized": false,
441
+ "special": true
442
+ },
443
+ "200053": {
444
+ "content": "<|unused_200053|>",
445
+ "single_word": false,
446
+ "lstrip": false,
447
+ "rstrip": false,
448
+ "normalized": false,
449
+ "special": true
450
+ },
451
+ "200054": {
452
+ "content": "<|unused_200054|>",
453
+ "single_word": false,
454
+ "lstrip": false,
455
+ "rstrip": false,
456
+ "normalized": false,
457
+ "special": true
458
+ },
459
+ "200055": {
460
+ "content": "<|unused_200055|>",
461
+ "single_word": false,
462
+ "lstrip": false,
463
+ "rstrip": false,
464
+ "normalized": false,
465
+ "special": true
466
+ },
467
+ "200056": {
468
+ "content": "<|unused_200056|>",
469
+ "single_word": false,
470
+ "lstrip": false,
471
+ "rstrip": false,
472
+ "normalized": false,
473
+ "special": true
474
+ },
475
+ "200057": {
476
+ "content": "<|content_invoke_tool_text|>",
477
+ "single_word": false,
478
+ "lstrip": false,
479
+ "rstrip": false,
480
+ "normalized": false,
481
+ "special": true
482
+ }
483
+ },
484
+ "clean_up_tokenization_spaces": false,
485
+ "fix_mistral_regex": false,
486
+ "extra_special_tokens": {
487
+ "endoftext": "<|endoftext|>",
488
+ "message_user": "<|message_user|>",
489
+ "message_model": "<|message_model|>",
490
+ "message_system": "<|message_system|>",
491
+ "message_tool": "<|message_tool|>",
492
+ "content_text": "<|content_text|>",
493
+ "content_image": "<|content_image|>",
494
+ "content_model_end_sampling": "<|content_model_end_sampling|>",
495
+ "content_thinking": "<|content_thinking|>",
496
+ "end_message": "<|end_message|>",
497
+ "content_audio_input": "<|content_audio_input|>",
498
+ "content_tool_error": "<|content_tool_error|>",
499
+ "audio": "<|audio|>",
500
+ "content_xml": "<|content_xml|>",
501
+ "begin_of_text": "<|begin_of_text|>",
502
+ "audio_end": "<|audio_end|>",
503
+ "content_invoke_tool_json": "<|content_invoke_tool_json|>",
504
+ "content_invoke_tool_text": "<|content_invoke_tool_text|>"
505
+ },
506
+ "model_max_length": 1000000000000000019884624838656,
507
+ "tokenizer_class": "PreTrainedTokenizerFast"
508
+ }