Update README.md
Browse files
README.md
CHANGED
|
@@ -75,6 +75,56 @@ Which might generate something like:
|
|
| 75 |
|
| 76 |
Same process applies. Usually, it is best to do a sliding window over the user and model turns, but keep the system prompt fixed at the start of the context window.
|
| 77 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 78 |
## Other notes
|
| 79 |
|
| 80 |
- When prompted correctly, the model will always start by generating a BOS token. This behavior is an accidental side-effect which we plan to address in future model versions and should not be relied upon.
|
|
|
|
| 75 |
|
| 76 |
Same process applies. Usually, it is best to do a sliding window over the user and model turns, but keep the system prompt fixed at the start of the context window.
|
| 77 |
|
| 78 |
+
## Eval / Benchmark scores
|
| 79 |
+
|
| 80 |
+
Current eval out of the current 16bit Pygmalion-13b model: <br>
|
| 81 |
+
<html>
|
| 82 |
+
<head>
|
| 83 |
+
<style>
|
| 84 |
+
table {
|
| 85 |
+
border:1px solid #b3adad;
|
| 86 |
+
border-collapse:collapse;
|
| 87 |
+
padding:5px;
|
| 88 |
+
}
|
| 89 |
+
table th {
|
| 90 |
+
border:1px solid #b3adad;
|
| 91 |
+
padding:5px;
|
| 92 |
+
background: #f0f0f0;
|
| 93 |
+
color: #313030;
|
| 94 |
+
}
|
| 95 |
+
table td {
|
| 96 |
+
border:1px solid #b3adad;
|
| 97 |
+
text-align:center;
|
| 98 |
+
padding:5px;
|
| 99 |
+
background: #ffffff;
|
| 100 |
+
color: #313030;
|
| 101 |
+
}
|
| 102 |
+
</style>
|
| 103 |
+
</head>
|
| 104 |
+
<body>
|
| 105 |
+
<table>
|
| 106 |
+
<thead>
|
| 107 |
+
<tr>
|
| 108 |
+
<th>Model:</th>
|
| 109 |
+
<th>Wikitext2</th>
|
| 110 |
+
<th>Ptb-New</th>
|
| 111 |
+
<th>C4-New</th>
|
| 112 |
+
</tr>
|
| 113 |
+
</thead>
|
| 114 |
+
<tbody>
|
| 115 |
+
<tr>
|
| 116 |
+
<td>Metharme 13b - 16bit</td>
|
| 117 |
+
<td>5.253076553344727</td>
|
| 118 |
+
<td>27.53407859802246</td>
|
| 119 |
+
<td>TBD</td>
|
| 120 |
+
</tr>
|
| 121 |
+
</tbody>
|
| 122 |
+
</table>
|
| 123 |
+
</body>
|
| 124 |
+
</html>
|
| 125 |
+
|
| 126 |
+
<hr>
|
| 127 |
+
|
| 128 |
## Other notes
|
| 129 |
|
| 130 |
- When prompted correctly, the model will always start by generating a BOS token. This behavior is an accidental side-effect which we plan to address in future model versions and should not be relied upon.
|