Hugging Face's logo

inclusionAI
/
Ling-3.0-tiny-base

Safetensors
bailing_hybrid
custom_code
Model card Files
xet
 Community
Ling-3.0-tiny-base
File size: 17,013 Bytes
7da73ee
 
 
96444ff
7da73ee
 
 
 
 
96444ff
bdd1bdf
7da73ee
bab7297
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7da73ee
 
96444ff
 
7da73ee
96444ff
 
 
7da73ee
96444ff
7da73ee
96444ff
 
bab7297
 
 
 
7da73ee
 
bab7297
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7da73ee
bab7297
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7da73ee
96444ff
0799d31
7da73ee
f141aa7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7da73ee
96444ff
 
7da73ee
96444ff
 
 
 
 
7da73ee
96444ff
7da73ee
96444ff
 
 
7da73ee
96444ff
 
7da73ee
96444ff
 
7da73ee
96444ff
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
---
license: mit
---

<p align="center">
    <img src="https://mdn.alipayobjects.com/huamei_qa8qxu/afts/img/A*4QxcQrBlTiAAAAAAQXAAAAgAemJ7AQ/original" width="100"/>
</p>
<p align="center">🤗 <a href="https://huggingface.co/inclusionAI">Hugging Face</a>&nbsp;&nbsp; | &nbsp;&nbsp;🤖 <a href="https://modelscope.cn/organization/inclusionAI">ModelScope </a>&nbsp;&nbsp; | &nbsp;&nbsp;🐙 <a href="https://openrouter.ai/inclusionai/ling-3.0-tiny:free">OpenRouter </a>&nbsp;&nbsp;</p>

## Introduction
We have open-sourced the Ling-3.0 series, our most efficient language foundation model family to date. To support research and community-driven innovation, we are releasing **a collection of checkpoints during the training process** as following:

<div style="width: 100%; overflow-x: auto;">
  <table
    align="center"
    style="margin: 0 auto; text-align: center; border-collapse: collapse;"
  >
    <thead>
      <tr>
        <th align="center">Model</th>
        <th align="center">Pre-trained</th>
        <th align="center">Mid-trained</th>
        <th align="center">
          Merged (i.e., <a href="https://arxiv.org/abs/2507.17634">WSM</a>)
        </th>
      </tr>
    </thead>
    <tbody>
      <tr>
        <td align="center">Ling-3.0-tiny</td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-tiny-base-30T">
            <strong>Ling-3.0-tiny-base-30T</strong>
          </a>
        </td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-tiny-base-midtrain">
            <strong>Ling-3.0-tiny-base-midtrain</strong>
          </a>
        </td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-tiny-base">
            <strong>Ling-3.0-tiny-base</strong>
          </a>
        </td>
      </tr>
      <tr>
        <td align="center">Ling-3.0-flash</td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-flash-base-30T">
            <strong>Ling-3.0-flash-base-30T</strong>
          </a>
        </td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-flash-base-midtrain">
            <strong>Ling-3.0-flash-base-midtrain</strong>
          </a>
        </td>
        <td align="center">
          <a href="https://huggingface.co/inclusionAI/Ling-3.0-flash-base">
            <strong>Ling-3.0-flash-base</strong>
          </a>
        </td>
      </tr>
    </tbody>
  </table>
</div>


  
These checkpoints correspond to different stages of the training process:

+ **Pretrained checkpoint** have completed large-scale pretraining but have not undergone mid-training, WSM merging (or learning-rate decay), or post-training.
+ **Mid-trained checkpoint** have completed mid-training but have not undergone WSM merging (or learning-rate decay) or post-training.
+ **Merged checkpoints** have undergone [WSM](https://arxiv.org/abs/2507.17634) merging (or learning-rate decay) based on the mid-training checkpoints but have not undergone post-training.

These checkpoints are released to support continued pretraining, fine-tuning, and further research. For the post-trained model, please see [Ling-3.0-tiny](https://huggingface.co/inclusionAI/Ling-3.0-tiny) and [Ling-3.0-flash](https://huggingface.co/inclusionAI/Ling-3.0-flash).

## Model Overview
### Key features
+ **Highly sparse (1/64) MoE architecture**: 512 routed experts, with only 8 routed experts and 1 shared expert activated per token. This enables broad model capabilities while activating just 5.1B (Non-emb) parameters per token;
+ **Native hybrid linear attention**: Ling-3.0 series adopt a native hybrid linear attention architecture from the very start of pretraining by combining KDA with Gated MLA to enable efficient processing of long-context inputs.
+ [**Warmup-Stable and Merge**](https://arxiv.org/abs/2507.17634): We replace conventional learning-rate decay with weighted checkpoint merging. By eliminating the decay phase, our Base Model is better suited for continual pretraining and dynamic data expansion, while enabling offline exploration of different decay profiles without rerunning costly experiments for each strategy.
+ **Scale Seamlessly**: Ling-3.0-tiny-base and Ling-3.0-flash-base share the same training recipe, enabling community to experiment on the Ling-3.0-tiny-base first and then scale validated training strategies to the larger Ling-3.0-flash-base.


<div
  style="
    display: flex;
    align-items: center;
    justify-content: center;
    gap: 32px;
    width: 100%;
    max-width: 1200px;
    margin: 0 auto;
    flex-wrap: wrap;
  "
>
  <div style="flex: 1 1 420px; max-width: 560px; overflow-x: auto;">
    <table style="width: 100%; text-align: center; border-collapse: collapse;">
      <thead>
        <tr>
          <th align="center">Model Type</th>
          <th align="center">Base (final checkpoint of WSM merging)</th>
        </tr>
      </thead>
      <tbody>
        <tr><td align="center">Architecture</td><td align="center">Hybrid-linear MoE</td></tr>
        <tr><td align="center">Parameter Scale</td><td align="center">Totoal 7.9B, Activated 1.3B</td></tr>
        <tr><td align="center">Transformer Layers</td><td align="center">18 KDA + 6 Gated MLA (3:1)</td></tr>
        <tr><td align="center">Number of Dense Layers</td><td align="center">1</td></tr>
        <tr><td align="center">Number of Routed Experts</td><td align="center">128</td></tr>
        <tr><td align="center">Number of Shared Experts</td><td align="center">1</td></tr>
        <tr><td align="center">Number of Activated Experts</td><td align="center">8</td></tr>
        <tr><td align="center">Attention Heads</td><td align="center">16</td></tr>
        <tr><td align="center">Hidden Size</td><td align="center">1536</td></tr>
        <tr><td align="center">Expert Intermediate Size</td><td align="center">512</td></tr>
        <tr><td align="center">Dense Intermediate Size</td><td align="center">4608</td></tr>
        <tr><td align="center">Vocabulary Size</td><td align="center">157,184</td></tr>
      </tbody>
    </table>
  </div>

  <div style="flex: 0 1 700px; max-width: 800px;">
    <img
      src="https://intranetproxy.alipay.com/skylark/lark/0/2026/png/62256938/1787120625910-bb0c32bd-7e6b-4354-ac65-fa5b747d8bff.png"
      width="520"
      style="
        display: block;
        width: 100%;
        max-width: 800px;
        height: auto;
        margin: 0 auto;
        border-radius: 10px;
      "
      alt=""
    >
  </div>
</div>

## Base Model Evaluation
To systematically assess the capabilities of the base model, we use a self-built comprehensive benchmark suite covering several key domains, including knowledge, coding, mathematics, reasoning, multilingual understanding, and long-context comprehension. The performance of the pretrained base checkpoint, i.e., `Ling-3.0-tiny-base`, is compared below:

<style>
  .benchmark-table {
    border-collapse: collapse;
    margin: 0 auto;
  }

  .benchmark-table th,
  .benchmark-table td {
    padding: 0.55rem 0.8rem;
    text-align: center !important;
    vertical-align: middle;
  }

  .benchmark-table .ling {
    background-color: #e7f6ff;
  }

  .benchmark-table thead .ling {
    color: #1677ff;
  }

  .benchmark-table thead tr:first-child .ling {
    border-radius: 28px 28px 0 0;
  }

  .benchmark-table tbody tr:last-child .ling {
    border-radius: 0 0 28px 28px;
  }

  .benchmark-table .section-start > * {
    border-top: 2px solid #d0d7de;
  }

  .benchmark-table ins {
    text-decoration-thickness: 1px;
    text-underline-offset: 0.12em;
  }

  .benchmark-table .table-footnote {
    padding-top: 0.85rem;
    text-align: left !important;
    color: #57606a;
    font-size: 0.9em;
  }
</style>

<table class="benchmark-table" align="center">
  <thead>
    <tr align="center">
      <th align="center" rowspan="2">Domain</th>
      <th align="center" rowspan="2">Benchmark</th>
      <th align="center" rowspan="2">Shot Config</th>
      <th align="center" class="ling" bgcolor="#e7f6ff">Ling-3.0-tiny-base</th>
      <th align="center">Ling-2.5-mini-base</th>
      <th align="center">Qwen3.5-9B-base</th>
      <th align="center">Qwen3.5-4B-base</th>
    </tr>
    <tr align="center">
      <th align="center" class="ling" bgcolor="#e7f6ff">7.9B A1.3B</th>
      <th align="center">16B A1.4B</th>
      <th align="center">9B</th>
      <th align="center">4B</th>
    </tr>
  </thead>
  <tbody>
    <tr align="center">
      <th align="center" rowspan="6">Knowledge</th>
      <th align="center">CCPM<sub>(EM)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>84.01</ins></td>
      <td align="center">80.77</td>
      <td align="center"><strong>88.31</strong></td>
      <td align="center">81.88</td>
    </tr>
    <tr align="center">
      <th align="center">ARC-C<sub>(EM)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>92.20</ins></td>
      <td align="center">91.19</td>
      <td align="center"><strong>94.58</strong></td>
      <td align="center"><ins>92.20</ins></td>
    </tr>
    <tr align="center">
      <th align="center">AGIEval<sub>(Acc)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>64.49</ins></td>
      <td align="center">63.38</td>
      <td align="center"><strong>66.17</strong></td>
      <td align="center">61.45</td>
    </tr>
    <tr align="center">
      <th align="center">SimpleQA-Verified<sub>(Acc)</sub></th>
      <td align="center">5-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>7.20</ins></td>
      <td align="center">7.10</td>
      <td align="center"><strong>10.40</strong></td>
      <td align="center">6.40</td>
    </tr>
    <tr align="center">
      <th align="center">MMLU-Pro<sub>(EM)</sub></th>
      <td align="center">5-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff">51.83</td>
      <td align="center">49.89</td>
      <td align="center"><strong>57.76</strong></td>
      <td align="center"><ins>52.06</ins></td>
    </tr>
    <tr align="center">
      <th align="center">CEval<sub>(EM)</sub></th>
      <td align="center">5-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>80.59</ins></td>
      <td align="center">80.16</td>
      <td align="center"><strong>81.52</strong></td>
      <td align="center">76.67</td>
    </tr>
    <tr align="center" class="section-start">
      <th align="center" rowspan="7">Code</th>
      <th align="center">HumanEval-Plus<sub>(Pass@1)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>79.27</strong></td>
      <td align="center"><ins>76.22</ins></td>
      <td align="center">52.44</td>
      <td align="center">49.39</td>
    </tr>
    <tr align="center">
      <th align="center">CruxEval<sub>(Pass@1)</sub></th>
      <td align="center">1-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>67.44</ins></td>
      <td align="center">63.19</td>
      <td align="center"><strong>68.12</strong></td>
      <td align="center">64.31</td>
    </tr>
    <tr align="center">
      <th align="center">MultiPL-E<sub>(Pass@1)</sub></th>
      <td align="center">1-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>64.38</strong></td>
      <td align="center"><ins>64.34</ins></td>
      <td align="center">52.89</td>
      <td align="center">45.47</td>
    </tr>
    <tr align="center">
      <th align="center">LiveCodeBench<sup>1</sup><sub>(Pass@1)</sub></th>
      <td align="center">1-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>24.23</strong></td>
      <td align="center"><ins>22.91</ins></td>
      <td align="center">18.06</td>
      <td align="center">13.88</td>
    </tr>
    <tr align="center">
      <th align="center">BigCodeBench<sub>(Pass@1)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>42.89</strong></td>
      <td align="center"><ins>42.54</ins></td>
      <td align="center">29.39</td>
      <td align="center">24.39</td>
    </tr>
    <tr align="center">
      <th align="center">FullStackBench<sub>(Pass@1)</sub></th>
      <td align="center">3-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>39.48</strong></td>
      <td align="center"><ins>38.59</ins></td>
      <td align="center">37.23</td>
      <td align="center">33.14</td>
    </tr>
    <tr align="center">
      <th align="center">LCBench<sup>2</sup><sub>(Pass@1)</sub></th>
      <td align="center">3-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>41.59</ins></td>
      <td align="center"><strong>46.16</strong></td>
      <td align="center">33.47</td>
      <td align="center">19.46</td>
    </tr>
    <tr align="center" class="section-start">
      <th align="center" rowspan="4">Math</th>
      <th align="center">MATH500<sub>(Acc)</sub></th>
      <td align="center">4-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>65.60</ins></td>
      <td align="center"><strong>68.40</strong></td>
      <td align="center">55.20</td>
      <td align="center">49.60</td>
    </tr>
    <tr align="center">
      <th align="center">OlympiadBench<sub>(Acc)</sub></th>
      <td align="center">3-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>25.90</strong></td>
      <td align="center"><ins>24.10</ins></td>
      <td align="center">21.84</td>
      <td align="center">17.02</td>
    </tr>
    <tr align="center">
      <th align="center">TheoremQA<sub>(Acc)</sub></th>
      <td align="center">5-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>51.64</ins></td>
      <td align="center">50.68</td>
      <td align="center"><strong>52.88</strong></td>
      <td align="center">47.40</td>
    </tr>
    <tr align="center">
      <th align="center">OmniMath<sub>(Acc)</sub></th>
      <td align="center">3-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><strong>29.70</strong></td>
      <td align="center"><ins>29.02</ins></td>
      <td align="center">22.99</td>
      <td align="center">20.89</td>
    </tr>
    <tr align="center" class="section-start">
      <th align="center" rowspan="2">Reasoning</th>
      <th align="center">CommonSenseQA<sub>(EM)</sub></th>
      <td align="center">5-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>83.46</ins></td>
      <td align="center">81.16</td>
      <td align="center"><strong>83.62</strong></td>
      <td align="center">80.18</td>
    </tr>
    <tr align="center">
      <th align="center">BBH<sub>(EM)</sub></th>
      <td align="center">3-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>80.64</ins></td>
      <td align="center">77.37</td>
      <td align="center"><strong>84.52</strong></td>
      <td align="center">80.11</td>
    </tr>
    <tr align="center" class="section-start">
      <th align="center" rowspan="2">Long-context</th>
      <th align="center">LongBench<sub>(Acc)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>43.85</ins></td>
      <td align="center">29.93</td>
      <td align="center"><strong>51.87</strong></td>
      <td align="center">39.96</td>
    </tr>
    <tr align="center">
      <th align="center">LEval<sub>(Acc)</sub></th>
      <td align="center">0-shot</td>
      <td align="center" class="ling" bgcolor="#e7f6ff"><ins>68.37</ins></td>
      <td align="center">62.72</td>
      <td align="center"><strong>77.24</strong></td>
      <td align="center">63.09</td>
    </tr>
  </tbody>
  <tfoot>
    <tr>
      <td class="table-footnote" colspan="7" align="left">
        <strong>Note:</strong><br>
        <sup>1</sup> LiveCodeBench (2408-2505)<br>
        <sup>2</sup> LCBench (2301-2502)
      </td>
    </tr>
  </tfoot>
</table>



<!-- <img src="https://intranetproxy.alipay.com/skylark/lark/0/2026/png/62256938/1787145422924-5c03d4b9-ee56-4d84-8927-58edb24b24c1.png" width="1578.5" title="" crop="0,0,1,1" id="u9090be40" class="ne-image"> -->

## **Intended Use**
Recommended use cases: 

+ Continued pre-training 
+ Mid-training
+ Supervised fine-tuning for domain adaptation 
+ Preference optimization and RL post-training Distillation research 
+ Long-context and MoE systems research 

Not recommended as-is for: 

+ Direct end-user chat deployment 
+ Safety-critical applications without additional alignment and evaluation 
+ Production use without post-training and task-specific validation

## Usage
For fine-tuning examples, please refer to our [ling-cookbook](https://github.com/inclusionAI/ling-cookbook/).

## FAQ
If you have any question, please feel free to add a discussion.

## License
This model is released under the **MIT License**.