Commit
•
038bd78
1
Parent(s):
4d9a9de
Update README.md (#1)
Browse files- Update README.md (5aa3a0997fb4daf67960a8917a51cc5f041ebfd4)
Co-authored-by: Qiyang Zhao <[email protected]>
README.md
CHANGED
|
@@ -71,8 +71,61 @@ python run_inference.py -m models/Falcon3-1B-1.58bit/ggml-model-i2_s.gguf -p "Yo
|
|
| 71 |
```
|
| 72 |
|
| 73 |
# Evaluation
|
| 74 |
-
|
| 75 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
|
| 77 |
## Useful links
|
| 78 |
- View our [release blogpost](https://huggingface.co/blog/falcon3).
|
|
|
|
| 71 |
```
|
| 72 |
|
| 73 |
# Evaluation
|
| 74 |
+
We report in the following table our internal pipeline benchmarks:
|
| 75 |
+
|
| 76 |
+
**Note evaluation results are normalized score from v2 leaderboard tasks - reported results of original models in the blogpost are raw scores**
|
| 77 |
+
|
| 78 |
+
<table border="1" style="width: 100%; text-align: center; border-collapse: collapse;">
|
| 79 |
+
<colgroup>
|
| 80 |
+
<col style="width: 10%;">
|
| 81 |
+
<col style="width: 10%;">
|
| 82 |
+
<col style="background-color: rgba(80, 15, 213, 0.5); width: 7%;">
|
| 83 |
+
</colgroup>
|
| 84 |
+
<thead>
|
| 85 |
+
<tr>
|
| 86 |
+
<th>Benchmark</th>
|
| 87 |
+
<th>Llama3-8B-1.58-100B-tokens</th>
|
| 88 |
+
<th>Falcon3-1B-Instruct-1.58bit</th>
|
| 89 |
+
</tr>
|
| 90 |
+
</thead>
|
| 91 |
+
<tbody>
|
| 92 |
+
<tr>
|
| 93 |
+
<td>IFEval</td>
|
| 94 |
+
<td>17.91</td>
|
| 95 |
+
<td><b>44.5</b></td>
|
| 96 |
+
</tr>
|
| 97 |
+
<tr>
|
| 98 |
+
<td>MUSR</td>
|
| 99 |
+
<td>4.87</td>
|
| 100 |
+
<td><b>2.78</b></td>
|
| 101 |
+
</tr>
|
| 102 |
+
<tr>
|
| 103 |
+
<td>GPQA</td>
|
| 104 |
+
<td>1.83</td>
|
| 105 |
+
<td><b>0</b></td>
|
| 106 |
+
</tr>
|
| 107 |
+
<tr>
|
| 108 |
+
<td>BBH</td>
|
| 109 |
+
<td><b>5.36</b></td>
|
| 110 |
+
<td>2.24</td>
|
| 111 |
+
</tr>
|
| 112 |
+
<tr>
|
| 113 |
+
<td>MMLU-PRO</td>
|
| 114 |
+
<td><b>2.78</b></td>
|
| 115 |
+
<td>1.93</td>
|
| 116 |
+
</tr>
|
| 117 |
+
<tr>
|
| 118 |
+
<td>MATH</td>
|
| 119 |
+
<td>0.26</td>
|
| 120 |
+
<td><b>0.17</b></td>
|
| 121 |
+
</tr>
|
| 122 |
+
<tr>
|
| 123 |
+
<td>Average</td>
|
| 124 |
+
<td>5.5</td>
|
| 125 |
+
<td><b>8.6</b></td>
|
| 126 |
+
</tr>
|
| 127 |
+
</tbody>
|
| 128 |
+
</table>
|
| 129 |
|
| 130 |
## Useful links
|
| 131 |
- View our [release blogpost](https://huggingface.co/blog/falcon3).
|