Update README.md
Browse files
README.md
CHANGED
|
@@ -45,6 +45,100 @@ We have released the results for LoRAs and LoRA-Flow
|
|
| 45 |
| | LoRA-Hub | 20.8 | 28.4 | 36.8 | 28.7 | 19.5 | 21.3 | 20.1 | 20.3 |
|
| 46 |
| | LoRA-Flow | **33.2** | **37.6**| **42.0**| **37.6**| **20.7** | **23.8**| **23.2**| **22.6**|
|
| 47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
|
| 49 |
# Citation
|
| 50 |
```bibtex
|
|
|
|
| 45 |
| | LoRA-Hub | 20.8 | 28.4 | 36.8 | 28.7 | 19.5 | 21.3 | 20.1 | 20.3 |
|
| 46 |
| | LoRA-Flow | **33.2** | **37.6**| **42.0**| **37.6**| **20.7** | **23.8**| **23.2**| **22.6**|
|
| 47 |
|
| 48 |
+
<table border="1" cellspacing="0" cellpadding="10">
|
| 49 |
+
<caption>
|
| 50 |
+
Evaluation results on MGSM and HumanEval. ‘Lang’ denotes the chat LoRA in the target language and ‘Task’ represents the math or code LoRA trained in English. LoRA fusion methods combine the language LoRA and the task LoRA to accomplish the new task. The best score is highlighted in bold.
|
| 51 |
+
</caption>
|
| 52 |
+
<thead>
|
| 53 |
+
<tr>
|
| 54 |
+
<th rowspan="2">Method</th>
|
| 55 |
+
<th colspan="4">MGSM (Math)</th>
|
| 56 |
+
<th colspan="4">HumanEval (Code)</th>
|
| 57 |
+
</tr>
|
| 58 |
+
<tr>
|
| 59 |
+
<th>Zh</th>
|
| 60 |
+
<th>Ru</th>
|
| 61 |
+
<th>Es</th>
|
| 62 |
+
<th>Avg.</th>
|
| 63 |
+
<th>Zh</th>
|
| 64 |
+
<th>Ru</th>
|
| 65 |
+
<th>Es</th>
|
| 66 |
+
<th>Avg.</th>
|
| 67 |
+
</tr>
|
| 68 |
+
</thead>
|
| 69 |
+
<tbody>
|
| 70 |
+
<tr>
|
| 71 |
+
<td colspan="2">Base Model</td>
|
| 72 |
+
<td>4.4</td>
|
| 73 |
+
<td>3.2</td>
|
| 74 |
+
<td>2.4</td>
|
| 75 |
+
<td>3.3</td>
|
| 76 |
+
<td>0.0</td>
|
| 77 |
+
<td>0.0</td>
|
| 78 |
+
<td>2.4</td>
|
| 79 |
+
<td>0.8</td>
|
| 80 |
+
</tr>
|
| 81 |
+
<tr>
|
| 82 |
+
<td rowspan="2">Single LoRA</td>
|
| 83 |
+
<td>Lang</td>
|
| 84 |
+
<td>5.2</td>
|
| 85 |
+
<td>3.6</td>
|
| 86 |
+
<td>3.6</td>
|
| 87 |
+
<td>4.1</td>
|
| 88 |
+
<td>12.2</td>
|
| 89 |
+
<td>14.0</td>
|
| 90 |
+
<td>10.4</td>
|
| 91 |
+
<td>12.2</td>
|
| 92 |
+
</tr>
|
| 93 |
+
<tr>
|
| 94 |
+
<td>Task</td>
|
| 95 |
+
<td>26.8</td>
|
| 96 |
+
<td>32.8</td>
|
| 97 |
+
<td>41.2</td>
|
| 98 |
+
<td>33.6</td>
|
| 99 |
+
<td>18.3</td>
|
| 100 |
+
<td>23.2</td>
|
| 101 |
+
<td>21.9</td>
|
| 102 |
+
<td>21.1</td>
|
| 103 |
+
</tr>
|
| 104 |
+
<tr>
|
| 105 |
+
<td rowspan="3">LoRA Fusion</td>
|
| 106 |
+
<td>Average</td>
|
| 107 |
+
<td>12.8</td>
|
| 108 |
+
<td>10.4</td>
|
| 109 |
+
<td>18.4</td>
|
| 110 |
+
<td>13.9</td>
|
| 111 |
+
<td>17.1</td>
|
| 112 |
+
<td>17.7</td>
|
| 113 |
+
<td>18.3</td>
|
| 114 |
+
<td>17.7</td>
|
| 115 |
+
</tr>
|
| 116 |
+
<tr>
|
| 117 |
+
<td>LoRA-Hub</td>
|
| 118 |
+
<td>20.8</td>
|
| 119 |
+
<td>28.4</td>
|
| 120 |
+
<td>36.8</td>
|
| 121 |
+
<td>28.7</td>
|
| 122 |
+
<td>19.5</td>
|
| 123 |
+
<td>21.3</td>
|
| 124 |
+
<td>20.1</td>
|
| 125 |
+
<td>20.3</td>
|
| 126 |
+
</tr>
|
| 127 |
+
<tr>
|
| 128 |
+
<td>LoRA-Flow</td>
|
| 129 |
+
<td><strong>33.2</strong></td>
|
| 130 |
+
<td><strong>37.6</strong></td>
|
| 131 |
+
<td><strong>42.0</strong></td>
|
| 132 |
+
<td><strong>37.6</strong></td>
|
| 133 |
+
<td><strong>20.7</strong></td>
|
| 134 |
+
<td><strong>23.8</strong></td>
|
| 135 |
+
<td><strong>23.2</strong></td>
|
| 136 |
+
<td><strong>22.6</strong></td>
|
| 137 |
+
</tr>
|
| 138 |
+
</tbody>
|
| 139 |
+
</table>
|
| 140 |
+
|
| 141 |
+
|
| 142 |
|
| 143 |
# Citation
|
| 144 |
```bibtex
|