Update README.md
Browse files
README.md
CHANGED
|
@@ -54,7 +54,259 @@ These checkpoints are released to support continued pretraining, fine-tuning, an
|
|
| 54 |
## Base Model Evaluation
|
| 55 |
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:
|
| 56 |
|
| 57 |
-
<
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
## **Intended Use**
|
| 60 |
Recommended use cases:
|
|
|
|
| 54 |
## Base Model Evaluation
|
| 55 |
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:
|
| 56 |
|
| 57 |
+
<style>
|
| 58 |
+
.benchmark-table {
|
| 59 |
+
border-collapse: collapse;
|
| 60 |
+
margin: 0 auto;
|
| 61 |
+
}
|
| 62 |
+
|
| 63 |
+
.benchmark-table th,
|
| 64 |
+
.benchmark-table td {
|
| 65 |
+
padding: 0.55rem 0.8rem;
|
| 66 |
+
text-align: center !important;
|
| 67 |
+
vertical-align: middle;
|
| 68 |
+
}
|
| 69 |
+
|
| 70 |
+
.benchmark-table .ling {
|
| 71 |
+
background-color: #e7f6ff;
|
| 72 |
+
}
|
| 73 |
+
|
| 74 |
+
.benchmark-table thead .ling {
|
| 75 |
+
color: #1677ff;
|
| 76 |
+
}
|
| 77 |
+
|
| 78 |
+
.benchmark-table thead tr:first-child .ling {
|
| 79 |
+
border-radius: 28px 28px 0 0;
|
| 80 |
+
}
|
| 81 |
+
|
| 82 |
+
.benchmark-table tbody tr:last-child .ling {
|
| 83 |
+
border-radius: 0 0 28px 28px;
|
| 84 |
+
}
|
| 85 |
+
|
| 86 |
+
.benchmark-table .section-start > * {
|
| 87 |
+
border-top: 2px solid #d0d7de;
|
| 88 |
+
}
|
| 89 |
+
|
| 90 |
+
.benchmark-table ins {
|
| 91 |
+
text-decoration-thickness: 1px;
|
| 92 |
+
text-underline-offset: 0.12em;
|
| 93 |
+
}
|
| 94 |
+
|
| 95 |
+
.benchmark-table .table-footnote {
|
| 96 |
+
padding-top: 0.85rem;
|
| 97 |
+
text-align: left !important;
|
| 98 |
+
color: #57606a;
|
| 99 |
+
font-size: 0.9em;
|
| 100 |
+
}
|
| 101 |
+
</style>
|
| 102 |
+
|
| 103 |
+
<table class="benchmark-table" align="center">
|
| 104 |
+
<thead>
|
| 105 |
+
<tr align="center">
|
| 106 |
+
<th align="center" rowspan="2">Domain</th>
|
| 107 |
+
<th align="center" rowspan="2">Benchmark</th>
|
| 108 |
+
<th align="center" rowspan="2">Shot Config</th>
|
| 109 |
+
<th align="center" class="ling" bgcolor="#e7f6ff">Ling-3.0-tiny-base</th>
|
| 110 |
+
<th align="center">Ling-2.5-mini-base</th>
|
| 111 |
+
<th align="center">Qwen3.5-9B-base</th>
|
| 112 |
+
<th align="center">Qwen3.5-4B-base</th>
|
| 113 |
+
</tr>
|
| 114 |
+
<tr align="center">
|
| 115 |
+
<th align="center" class="ling" bgcolor="#e7f6ff">7.9B A1.3B</th>
|
| 116 |
+
<th align="center">16B A1.4B</th>
|
| 117 |
+
<th align="center">9B</th>
|
| 118 |
+
<th align="center">4B</th>
|
| 119 |
+
</tr>
|
| 120 |
+
</thead>
|
| 121 |
+
<tbody>
|
| 122 |
+
<tr align="center">
|
| 123 |
+
<th align="center" rowspan="6">Knowledge</th>
|
| 124 |
+
<th align="center">CCPM<sub>(EM)</sub></th>
|
| 125 |
+
<td align="center">0-shot</td>
|
| 126 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>84.01</ins></td>
|
| 127 |
+
<td align="center">80.77</td>
|
| 128 |
+
<td align="center"><strong>88.31</strong></td>
|
| 129 |
+
<td align="center">81.88</td>
|
| 130 |
+
</tr>
|
| 131 |
+
<tr align="center">
|
| 132 |
+
<th align="center">ARC-C<sub>(EM)</sub></th>
|
| 133 |
+
<td align="center">0-shot</td>
|
| 134 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>92.20</ins></td>
|
| 135 |
+
<td align="center">91.19</td>
|
| 136 |
+
<td align="center"><strong>94.58</strong></td>
|
| 137 |
+
<td align="center"><ins>92.20</ins></td>
|
| 138 |
+
</tr>
|
| 139 |
+
<tr align="center">
|
| 140 |
+
<th align="center">AGIEval<sub>(Acc)</sub></th>
|
| 141 |
+
<td align="center">0-shot</td>
|
| 142 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>64.49</ins></td>
|
| 143 |
+
<td align="center">63.38</td>
|
| 144 |
+
<td align="center"><strong>66.17</strong></td>
|
| 145 |
+
<td align="center">61.45</td>
|
| 146 |
+
</tr>
|
| 147 |
+
<tr align="center">
|
| 148 |
+
<th align="center">SimpleQA-Verified<sub>(Acc)</sub></th>
|
| 149 |
+
<td align="center">5-shot</td>
|
| 150 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>7.20</ins></td>
|
| 151 |
+
<td align="center">7.10</td>
|
| 152 |
+
<td align="center"><strong>10.40</strong></td>
|
| 153 |
+
<td align="center">6.40</td>
|
| 154 |
+
</tr>
|
| 155 |
+
<tr align="center">
|
| 156 |
+
<th align="center">MMLU-Pro<sub>(EM)</sub></th>
|
| 157 |
+
<td align="center">5-shot</td>
|
| 158 |
+
<td align="center" class="ling" bgcolor="#e7f6ff">51.83</td>
|
| 159 |
+
<td align="center">49.89</td>
|
| 160 |
+
<td align="center"><strong>57.76</strong></td>
|
| 161 |
+
<td align="center"><ins>52.06</ins></td>
|
| 162 |
+
</tr>
|
| 163 |
+
<tr align="center">
|
| 164 |
+
<th align="center">CEval<sub>(EM)</sub></th>
|
| 165 |
+
<td align="center">5-shot</td>
|
| 166 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>80.59</ins></td>
|
| 167 |
+
<td align="center">80.16</td>
|
| 168 |
+
<td align="center"><strong>81.52</strong></td>
|
| 169 |
+
<td align="center">76.67</td>
|
| 170 |
+
</tr>
|
| 171 |
+
<tr align="center" class="section-start">
|
| 172 |
+
<th align="center" rowspan="7">Code</th>
|
| 173 |
+
<th align="center">HumanEval-Plus<sub>(Pass@1)</sub></th>
|
| 174 |
+
<td align="center">0-shot</td>
|
| 175 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>79.27</strong></td>
|
| 176 |
+
<td align="center"><ins>76.22</ins></td>
|
| 177 |
+
<td align="center">52.44</td>
|
| 178 |
+
<td align="center">49.39</td>
|
| 179 |
+
</tr>
|
| 180 |
+
<tr align="center">
|
| 181 |
+
<th align="center">CruxEval<sub>(Pass@1)</sub></th>
|
| 182 |
+
<td align="center">1-shot</td>
|
| 183 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>67.44</ins></td>
|
| 184 |
+
<td align="center">63.19</td>
|
| 185 |
+
<td align="center"><strong>68.12</strong></td>
|
| 186 |
+
<td align="center">64.31</td>
|
| 187 |
+
</tr>
|
| 188 |
+
<tr align="center">
|
| 189 |
+
<th align="center">MultiPL-E<sub>(Pass@1)</sub></th>
|
| 190 |
+
<td align="center">1-shot</td>
|
| 191 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>64.38</strong></td>
|
| 192 |
+
<td align="center"><ins>64.34</ins></td>
|
| 193 |
+
<td align="center">52.89</td>
|
| 194 |
+
<td align="center">45.47</td>
|
| 195 |
+
</tr>
|
| 196 |
+
<tr align="center">
|
| 197 |
+
<th align="center">LiveCodeBench<sup>1</sup><sub>(Pass@1)</sub></th>
|
| 198 |
+
<td align="center">1-shot</td>
|
| 199 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>24.23</strong></td>
|
| 200 |
+
<td align="center"><ins>22.91</ins></td>
|
| 201 |
+
<td align="center">18.06</td>
|
| 202 |
+
<td align="center">13.88</td>
|
| 203 |
+
</tr>
|
| 204 |
+
<tr align="center">
|
| 205 |
+
<th align="center">BigCodeBench<sub>(Pass@1)</sub></th>
|
| 206 |
+
<td align="center">0-shot</td>
|
| 207 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>42.89</strong></td>
|
| 208 |
+
<td align="center"><ins>42.54</ins></td>
|
| 209 |
+
<td align="center">29.39</td>
|
| 210 |
+
<td align="center">24.39</td>
|
| 211 |
+
</tr>
|
| 212 |
+
<tr align="center">
|
| 213 |
+
<th align="center">FullStackBench<sub>(Pass@1)</sub></th>
|
| 214 |
+
<td align="center">3-shot</td>
|
| 215 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>39.48</strong></td>
|
| 216 |
+
<td align="center"><ins>38.59</ins></td>
|
| 217 |
+
<td align="center">37.23</td>
|
| 218 |
+
<td align="center">33.14</td>
|
| 219 |
+
</tr>
|
| 220 |
+
<tr align="center">
|
| 221 |
+
<th align="center">LCBench<sup>2</sup><sub>(Pass@1)</sub></th>
|
| 222 |
+
<td align="center">3-shot</td>
|
| 223 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>41.59</ins></td>
|
| 224 |
+
<td align="center"><strong>46.16</strong></td>
|
| 225 |
+
<td align="center">33.47</td>
|
| 226 |
+
<td align="center">19.46</td>
|
| 227 |
+
</tr>
|
| 228 |
+
<tr align="center" class="section-start">
|
| 229 |
+
<th align="center" rowspan="4">Math</th>
|
| 230 |
+
<th align="center">MATH500<sub>(Acc)</sub></th>
|
| 231 |
+
<td align="center">4-shot</td>
|
| 232 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>65.60</ins></td>
|
| 233 |
+
<td align="center"><strong>68.40</strong></td>
|
| 234 |
+
<td align="center">55.20</td>
|
| 235 |
+
<td align="center">49.60</td>
|
| 236 |
+
</tr>
|
| 237 |
+
<tr align="center">
|
| 238 |
+
<th align="center">OlympiadBench<sub>(Acc)</sub></th>
|
| 239 |
+
<td align="center">3-shot</td>
|
| 240 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>25.90</strong></td>
|
| 241 |
+
<td align="center"><ins>24.10</ins></td>
|
| 242 |
+
<td align="center">21.84</td>
|
| 243 |
+
<td align="center">17.02</td>
|
| 244 |
+
</tr>
|
| 245 |
+
<tr align="center">
|
| 246 |
+
<th align="center">TheoremQA<sub>(Acc)</sub></th>
|
| 247 |
+
<td align="center">5-shot</td>
|
| 248 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>51.64</ins></td>
|
| 249 |
+
<td align="center">50.68</td>
|
| 250 |
+
<td align="center"><strong>52.88</strong></td>
|
| 251 |
+
<td align="center">47.40</td>
|
| 252 |
+
</tr>
|
| 253 |
+
<tr align="center">
|
| 254 |
+
<th align="center">OmniMath<sub>(Acc)</sub></th>
|
| 255 |
+
<td align="center">3-shot</td>
|
| 256 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><strong>29.70</strong></td>
|
| 257 |
+
<td align="center"><ins>29.02</ins></td>
|
| 258 |
+
<td align="center">22.99</td>
|
| 259 |
+
<td align="center">20.89</td>
|
| 260 |
+
</tr>
|
| 261 |
+
<tr align="center" class="section-start">
|
| 262 |
+
<th align="center" rowspan="2">Reasoning</th>
|
| 263 |
+
<th align="center">CommonSenseQA<sub>(EM)</sub></th>
|
| 264 |
+
<td align="center">5-shot</td>
|
| 265 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>83.46</ins></td>
|
| 266 |
+
<td align="center">81.16</td>
|
| 267 |
+
<td align="center"><strong>83.62</strong></td>
|
| 268 |
+
<td align="center">80.18</td>
|
| 269 |
+
</tr>
|
| 270 |
+
<tr align="center">
|
| 271 |
+
<th align="center">BBH<sub>(EM)</sub></th>
|
| 272 |
+
<td align="center">3-shot</td>
|
| 273 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>80.64</ins></td>
|
| 274 |
+
<td align="center">77.37</td>
|
| 275 |
+
<td align="center"><strong>84.52</strong></td>
|
| 276 |
+
<td align="center">80.11</td>
|
| 277 |
+
</tr>
|
| 278 |
+
<tr align="center" class="section-start">
|
| 279 |
+
<th align="center" rowspan="2">Long-context</th>
|
| 280 |
+
<th align="center">LongBench<sub>(Acc)</sub></th>
|
| 281 |
+
<td align="center">0-shot</td>
|
| 282 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>43.85</ins></td>
|
| 283 |
+
<td align="center">29.93</td>
|
| 284 |
+
<td align="center"><strong>51.87</strong></td>
|
| 285 |
+
<td align="center">39.96</td>
|
| 286 |
+
</tr>
|
| 287 |
+
<tr align="center">
|
| 288 |
+
<th align="center">LEval<sub>(Acc)</sub></th>
|
| 289 |
+
<td align="center">0-shot</td>
|
| 290 |
+
<td align="center" class="ling" bgcolor="#e7f6ff"><ins>68.37</ins></td>
|
| 291 |
+
<td align="center">62.72</td>
|
| 292 |
+
<td align="center"><strong>77.24</strong></td>
|
| 293 |
+
<td align="center">63.09</td>
|
| 294 |
+
</tr>
|
| 295 |
+
</tbody>
|
| 296 |
+
<tfoot>
|
| 297 |
+
<tr>
|
| 298 |
+
<td class="table-footnote" colspan="7" align="left">
|
| 299 |
+
<strong>Note:</strong><br>
|
| 300 |
+
<sup>1</sup> LiveCodeBench (2408-2505)<br>
|
| 301 |
+
<sup>2</sup> LCBench (2301-2502)
|
| 302 |
+
</td>
|
| 303 |
+
</tr>
|
| 304 |
+
</tfoot>
|
| 305 |
+
</table>
|
| 306 |
+
|
| 307 |
+
|
| 308 |
+
|
| 309 |
+
<!-- <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"> -->
|
| 310 |
|
| 311 |
## **Intended Use**
|
| 312 |
Recommended use cases:
|