File size: 3,091 Bytes
1d6d955 a53584a 6f7802b a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a 6f7802b a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a c16fea1 a53584a | 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 | ---
license: apache-2.0
---
# NB-ROBERTA Training Code
This is the current training code for the planned nb-roberta models.
We are currently planning to run the following experiments:
<table>
<tr>
<td><strong>Name</strong>
</td>
<td><strong>nb-roberta-base-old (C)</strong>
</td>
</tr>
<tr>
<td>Corpus
</td>
<td>NbAiLab/nb_bert
</td>
</tr>
<tr>
<td>Pod size
</td>
<td>v4-64
</td>
</tr>
<tr>
<td>Batch size
</td>
<td>62*4*8 = 1984 = 2k
</td>
</tr>
<tr>
<td>Learning rate
</td>
<td>3e-4 (RoBERTa article is using 6e-4 and bs=8k)
</td>
</tr>
<tr>
<td>Number of steps
</td>
<td>250k
</td>
</tr>
</table>
<table>
<tr>
<td><strong>Name</strong>
</td>
<td><strong>nb-roberta-base-ext (B)</strong>
</td>
</tr>
<tr>
<td>Corpus
</td>
<td>NbAiLab/nbailab_extended
</td>
</tr>
<tr>
<td>Pod size
</td>
<td>v4-64
</td>
</tr>
<tr>
<td>Batch size
</td>
<td>62*4*8 = 1984 = 2k
</td>
</tr>
<tr>
<td>Learning rate
</td>
<td>3e-4 (RoBERTa article is using 6e-4 and bs=8k)
</td>
</tr>
<tr>
<td>Number of steps
</td>
<td>250k
</td>
</tr>
</table>
<table>
<tr>
<td><strong>Name</strong>
</td>
<td><strong>nb-roberta-large-ext</strong>
</td>
</tr>
<tr>
<td>Corpus
</td>
<td>NbAiLab/nbailab_extended
</td>
</tr>
<tr>
<td>Pod size
</td>
<td>v4-64
</td>
</tr>
<tr>
<td>Batch size
</td>
<td>32*4*8 = 2024 = 1k
</td>
</tr>
<tr>
<td>Learning rate
</td>
<td>2-e4 (RoBERTa article is using 4e-4 and bs=8k)
</td>
</tr>
<tr>
<td>Number of steps
</td>
<td>500k
</td>
</tr>
</table>
<table>
<tr>
<td><strong>Name</strong>
</td>
<td><strong>nb-roberta-base-scandi</strong>
</td>
</tr>
<tr>
<td>Corpus
</td>
<td>NbAiLab/scandinavian
</td>
</tr>
<tr>
<td>Pod size
</td>
<td>v4-64
</td>
</tr>
<tr>
<td>Batch size
</td>
<td>62*4*8 = 1984 = 2k
</td>
</tr>
<tr>
<td>Learning rate
</td>
<td>3e-4 (RoBERTa article is using 6e-4 and bs=8k)
</td>
</tr>
<tr>
<td>Number of steps
</td>
<td>250k
</td>
</tr>
</table>
<table>
<tr>
<td><strong>Name</strong>
</td>
<td><strong>nb-roberta-large-scandi</strong>
</td>
</tr>
<tr>
<td>Corpus
</td>
<td>NbAiLab/scandinavian
</td>
</tr>
<tr>
<td>Pod size
</td>
<td>v4-64
</td>
</tr>
<tr>
<td>Batch size
</td>
<td>32*4*8 = 1024 = 1k
</td>
</tr>
<tr>
<td>Learning rate
</td>
<td>2-e4 (RoBERTa article is using 4e-4 and bs=8k)
</td>
</tr>
<tr>
<td>Number of steps
</td>
<td>500k
</td>
</tr>
</table>
## Calculations
Some basic that we used when estimating the number of training steps:
* The Scandinavic Corpus is 85GB
* The Scandinavic Corpus contains 13B words
* With a conversion factor of 2.3, this is estimated to around 30B tokens
* 30B tokens / (512 seq length * 3000 batch size) = 20.000 steps
|