Datasets:
Adding boundaries to gaussian
Browse files- README.md +44 -300
- mc4-sampling.py +1 -1
README.md
CHANGED
|
@@ -135,33 +135,21 @@ task_ids:
|
|
| 135 |
paperswithcode_id: mc4
|
| 136 |
---
|
| 137 |
|
| 138 |
-
# Dataset Card for mC4
|
| 139 |
|
| 140 |
## Table of Contents
|
| 141 |
|
| 142 |
-
- [Dataset Card for mC4](#dataset-card-for-mc4)
|
| 143 |
- [Table of Contents](#table-of-contents)
|
| 144 |
- [Dataset Description](#dataset-description)
|
| 145 |
- [Dataset Summary](#dataset-summary)
|
|
|
|
| 146 |
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
| 147 |
- [Languages](#languages)
|
| 148 |
- [Dataset Structure](#dataset-structure)
|
| 149 |
- [Data Instances](#data-instances)
|
| 150 |
- [Data Fields](#data-fields)
|
| 151 |
- [Data Splits](#data-splits)
|
| 152 |
-
- [Dataset Creation](#dataset-creation)
|
| 153 |
-
- [Curation Rationale](#curation-rationale)
|
| 154 |
-
- [Source Data](#source-data)
|
| 155 |
-
- [Initial Data Collection and Normalization](#initial-data-collection-and-normalization)
|
| 156 |
-
- [Who are the source language producers?](#who-are-the-source-language-producers)
|
| 157 |
-
- [Annotations](#annotations)
|
| 158 |
-
- [Annotation process](#annotation-process)
|
| 159 |
-
- [Who are the annotators?](#who-are-the-annotators)
|
| 160 |
-
- [Personal and Sensitive Information](#personal-and-sensitive-information)
|
| 161 |
-
- [Considerations for Using the Data](#considerations-for-using-the-data)
|
| 162 |
-
- [Social Impact of Dataset](#social-impact-of-dataset)
|
| 163 |
-
- [Discussion of Biases](#discussion-of-biases)
|
| 164 |
-
- [Other Known Limitations](#other-known-limitations)
|
| 165 |
- [Additional Information](#additional-information)
|
| 166 |
- [Dataset Curators](#dataset-curators)
|
| 167 |
- [Licensing Information](#licensing-information)
|
|
@@ -170,131 +158,15 @@ paperswithcode_id: mc4
|
|
| 170 |
|
| 171 |
## Dataset Description
|
| 172 |
|
| 173 |
-
- **Homepage:** https://huggingface.co/
|
| 174 |
-
- **Paper:** https://arxiv.org/abs/1910.10683
|
| 175 |
|
| 176 |
### Dataset Summary
|
| 177 |
|
| 178 |
-
This dataset builds upon the original mC4 and adds
|
| 179 |
-
|
| 180 |
-
The original dataset is the multilingual colossal, cleaned version of Common Crawl's web crawl corpus. Based on Common Crawl dataset: "https://commoncrawl.org".
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
108 languages are available and are reported in the table below.
|
| 185 |
-
|
| 186 |
-
Note that the languages that end with "-Latn" are simply romanized variants, i.e. written using the Latin script.
|
| 187 |
-
|
| 188 |
-
| language code | language name |
|
| 189 |
-
|:----------------|:---------------------|
|
| 190 |
-
| af | Afrikaans |
|
| 191 |
-
| am | Amharic |
|
| 192 |
-
| ar | Arabic |
|
| 193 |
-
| az | Azerbaijani |
|
| 194 |
-
| be | Belarusian |
|
| 195 |
-
| bg | Bulgarian |
|
| 196 |
-
| bg-Latn | Bulgarian (Latin) |
|
| 197 |
-
| bn | Bangla |
|
| 198 |
-
| ca | Catalan |
|
| 199 |
-
| ceb | Cebuano |
|
| 200 |
-
| co | Corsican |
|
| 201 |
-
| cs | Czech |
|
| 202 |
-
| cy | Welsh |
|
| 203 |
-
| da | Danish |
|
| 204 |
-
| de | German |
|
| 205 |
-
| el | Greek |
|
| 206 |
-
| el-Latn | Greek (Latin) |
|
| 207 |
-
| en | English |
|
| 208 |
-
| eo | Esperanto |
|
| 209 |
-
| es | Spanish |
|
| 210 |
-
| et | Estonian |
|
| 211 |
-
| eu | Basque |
|
| 212 |
-
| fa | Persian |
|
| 213 |
-
| fi | Finnish |
|
| 214 |
-
| fil | Filipino |
|
| 215 |
-
| fr | French |
|
| 216 |
-
| fy | Western Frisian |
|
| 217 |
-
| ga | Irish |
|
| 218 |
-
| gd | Scottish Gaelic |
|
| 219 |
-
| gl | Galician |
|
| 220 |
-
| gu | Gujarati |
|
| 221 |
-
| ha | Hausa |
|
| 222 |
-
| haw | Hawaiian |
|
| 223 |
-
| hi | Hindi |
|
| 224 |
-
| hi-Latn | Hindi (Latin script) |
|
| 225 |
-
| hmn | Hmong, Mong |
|
| 226 |
-
| ht | Haitian |
|
| 227 |
-
| hu | Hungarian |
|
| 228 |
-
| hy | Armenian |
|
| 229 |
-
| id | Indonesian |
|
| 230 |
-
| ig | Igbo |
|
| 231 |
-
| is | Icelandic |
|
| 232 |
-
| it | Italian |
|
| 233 |
-
| iw | former Hebrew |
|
| 234 |
-
| ja | Japanese |
|
| 235 |
-
| ja-Latn | Japanese (Latin) |
|
| 236 |
-
| jv | Javanese |
|
| 237 |
-
| ka | Georgian |
|
| 238 |
-
| kk | Kazakh |
|
| 239 |
-
| km | Khmer |
|
| 240 |
-
| kn | Kannada |
|
| 241 |
-
| ko | Korean |
|
| 242 |
-
| ku | Kurdish |
|
| 243 |
-
| ky | Kyrgyz |
|
| 244 |
-
| la | Latin |
|
| 245 |
-
| lb | Luxembourgish |
|
| 246 |
-
| lo | Lao |
|
| 247 |
-
| lt | Lithuanian |
|
| 248 |
-
| lv | Latvian |
|
| 249 |
-
| mg | Malagasy |
|
| 250 |
-
| mi | Maori |
|
| 251 |
-
| mk | Macedonian |
|
| 252 |
-
| ml | Malayalam |
|
| 253 |
-
| mn | Mongolian |
|
| 254 |
-
| mr | Marathi |
|
| 255 |
-
| ms | Malay |
|
| 256 |
-
| mt | Maltese |
|
| 257 |
-
| my | Burmese |
|
| 258 |
-
| ne | Nepali |
|
| 259 |
-
| nl | Dutch |
|
| 260 |
-
| no | Norwegian |
|
| 261 |
-
| ny | Nyanja |
|
| 262 |
-
| pa | Punjabi |
|
| 263 |
-
| pl | Polish |
|
| 264 |
-
| ps | Pashto |
|
| 265 |
-
| pt | Portuguese |
|
| 266 |
-
| ro | Romanian |
|
| 267 |
-
| ru | Russian |
|
| 268 |
-
| ru-Latn | Russian (Latin) |
|
| 269 |
-
| sd | Sindhi |
|
| 270 |
-
| si | Sinhala |
|
| 271 |
-
| sk | Slovak |
|
| 272 |
-
| sl | Slovenian |
|
| 273 |
-
| sm | San Marino |
|
| 274 |
-
| sn | Shona |
|
| 275 |
-
| so | Somali |
|
| 276 |
-
| sq | Albanian |
|
| 277 |
-
| sr | Serbian |
|
| 278 |
-
| st | Southern Sotho |
|
| 279 |
-
| su | Sundanese |
|
| 280 |
-
| sv | Swedish |
|
| 281 |
-
| sw | Swahili |
|
| 282 |
-
| ta | Tamil |
|
| 283 |
-
| te | Telugu |
|
| 284 |
-
| tg | Tajik |
|
| 285 |
-
| th | Thai |
|
| 286 |
-
| tr | Turkish |
|
| 287 |
-
| uk | Ukrainian |
|
| 288 |
-
| und | Unknown language |
|
| 289 |
-
| ur | Urdu |
|
| 290 |
-
| uz | Uzbek |
|
| 291 |
-
| vi | Vietnamese |
|
| 292 |
-
| xh | Xhosa |
|
| 293 |
-
| yi | Yiddish |
|
| 294 |
-
| yo | Yoruba |
|
| 295 |
-
| zh | Chinese |
|
| 296 |
-
| zh-Latn | Chinese (Latin) |
|
| 297 |
-
| zu | Zulu |
|
| 298 |
|
| 299 |
You can load the mC4 subset of any language like this:
|
| 300 |
|
|
@@ -312,9 +184,40 @@ from datasets import load_dataset
|
|
| 312 |
mc4_subset_with_five_languages = load_dataset("mc4", languages=["en", "fr", "es", "de", "zh"])
|
| 313 |
```
|
| 314 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 315 |
### Supported Tasks and Leaderboards
|
| 316 |
|
| 317 |
-
mC4 is mainly intended to pretrain language models and word representations.
|
| 318 |
|
| 319 |
### Languages
|
| 320 |
|
|
@@ -357,176 +260,17 @@ The data have several fields:
|
|
| 357 |
|
| 358 |
### Data Splits
|
| 359 |
|
| 360 |
-
|
| 361 |
-
|
| 362 |
-
| config | train | validation |
|
| 363 |
-
|:---------|:--------|:-------------|
|
| 364 |
-
| af | ? | ? |
|
| 365 |
-
| am | ? | ? |
|
| 366 |
-
| ar | ? | ? |
|
| 367 |
-
| az | ? | ? |
|
| 368 |
-
| be | ? | ? |
|
| 369 |
-
| bg | ? | ? |
|
| 370 |
-
| bg-Latn | ? | ? |
|
| 371 |
-
| bn | ? | ? |
|
| 372 |
-
| ca | ? | ? |
|
| 373 |
-
| ceb | ? | ? |
|
| 374 |
-
| co | ? | ? |
|
| 375 |
-
| cs | ? | ? |
|
| 376 |
-
| cy | ? | ? |
|
| 377 |
-
| da | ? | ? |
|
| 378 |
-
| de | ? | ? |
|
| 379 |
-
| el | ? | ? |
|
| 380 |
-
| el-Latn | ? | ? |
|
| 381 |
-
| en | ? | ? |
|
| 382 |
-
| eo | ? | ? |
|
| 383 |
-
| es | ? | ? |
|
| 384 |
-
| et | ? | ? |
|
| 385 |
-
| eu | ? | ? |
|
| 386 |
-
| fa | ? | ? |
|
| 387 |
-
| fi | ? | ? |
|
| 388 |
-
| fil | ? | ? |
|
| 389 |
-
| fr | ? | ? |
|
| 390 |
-
| fy | ? | ? |
|
| 391 |
-
| ga | ? | ? |
|
| 392 |
-
| gd | ? | ? |
|
| 393 |
-
| gl | ? | ? |
|
| 394 |
-
| gu | ? | ? |
|
| 395 |
-
| ha | ? | ? |
|
| 396 |
-
| haw | ? | ? |
|
| 397 |
-
| hi | ? | ? |
|
| 398 |
-
| hi-Latn | ? | ? |
|
| 399 |
-
| hmn | ? | ? |
|
| 400 |
-
| ht | ? | ? |
|
| 401 |
-
| hu | ? | ? |
|
| 402 |
-
| hy | ? | ? |
|
| 403 |
-
| id | ? | ? |
|
| 404 |
-
| ig | ? | ? |
|
| 405 |
-
| is | ? | ? |
|
| 406 |
-
| it | ? | ? |
|
| 407 |
-
| iw | ? | ? |
|
| 408 |
-
| ja | ? | ? |
|
| 409 |
-
| ja-Latn | ? | ? |
|
| 410 |
-
| jv | ? | ? |
|
| 411 |
-
| ka | ? | ? |
|
| 412 |
-
| kk | ? | ? |
|
| 413 |
-
| km | ? | ? |
|
| 414 |
-
| kn | ? | ? |
|
| 415 |
-
| ko | ? | ? |
|
| 416 |
-
| ku | ? | ? |
|
| 417 |
-
| ky | ? | ? |
|
| 418 |
-
| la | ? | ? |
|
| 419 |
-
| lb | ? | ? |
|
| 420 |
-
| lo | ? | ? |
|
| 421 |
-
| lt | ? | ? |
|
| 422 |
-
| lv | ? | ? |
|
| 423 |
-
| mg | ? | ? |
|
| 424 |
-
| mi | ? | ? |
|
| 425 |
-
| mk | ? | ? |
|
| 426 |
-
| ml | ? | ? |
|
| 427 |
-
| mn | ? | ? |
|
| 428 |
-
| mr | ? | ? |
|
| 429 |
-
| ms | ? | ? |
|
| 430 |
-
| mt | ? | ? |
|
| 431 |
-
| my | ? | ? |
|
| 432 |
-
| ne | ? | ? |
|
| 433 |
-
| nl | ? | ? |
|
| 434 |
-
| no | ? | ? |
|
| 435 |
-
| ny | ? | ? |
|
| 436 |
-
| pa | ? | ? |
|
| 437 |
-
| pl | ? | ? |
|
| 438 |
-
| ps | ? | ? |
|
| 439 |
-
| pt | ? | ? |
|
| 440 |
-
| ro | ? | ? |
|
| 441 |
-
| ru | ? | ? |
|
| 442 |
-
| ru-Latn | ? | ? |
|
| 443 |
-
| sd | ? | ? |
|
| 444 |
-
| si | ? | ? |
|
| 445 |
-
| sk | ? | ? |
|
| 446 |
-
| sl | ? | ? |
|
| 447 |
-
| sm | ? | ? |
|
| 448 |
-
| sn | ? | ? |
|
| 449 |
-
| so | ? | ? |
|
| 450 |
-
| sq | ? | ? |
|
| 451 |
-
| sr | ? | ? |
|
| 452 |
-
| st | ? | ? |
|
| 453 |
-
| su | ? | ? |
|
| 454 |
-
| sv | ? | ? |
|
| 455 |
-
| sw | ? | ? |
|
| 456 |
-
| ta | ? | ? |
|
| 457 |
-
| te | ? | ? |
|
| 458 |
-
| tg | ? | ? |
|
| 459 |
-
| th | ? | ? |
|
| 460 |
-
| tr | ? | ? |
|
| 461 |
-
| uk | ? | ? |
|
| 462 |
-
| und | ? | ? |
|
| 463 |
-
| ur | ? | ? |
|
| 464 |
-
| uz | ? | ? |
|
| 465 |
-
| vi | ? | ? |
|
| 466 |
-
| xh | ? | ? |
|
| 467 |
-
| yi | ? | ? |
|
| 468 |
-
| yo | ? | ? |
|
| 469 |
-
| zh | ? | ? |
|
| 470 |
-
| zh-Latn | ? | ? |
|
| 471 |
-
| zu | ? | ? |
|
| 472 |
-
|
| 473 |
-
## Dataset Creation
|
| 474 |
-
|
| 475 |
-
### Curation Rationale
|
| 476 |
-
|
| 477 |
-
[More Information Needed]
|
| 478 |
-
|
| 479 |
-
### Source Data
|
| 480 |
-
|
| 481 |
-
#### Initial Data Collection and Normalization
|
| 482 |
-
|
| 483 |
-
[More Information Needed]
|
| 484 |
-
|
| 485 |
-
#### Who are the source language producers?
|
| 486 |
-
|
| 487 |
-
[More Information Needed]
|
| 488 |
-
|
| 489 |
-
### Annotations
|
| 490 |
-
|
| 491 |
-
#### Annotation process
|
| 492 |
-
|
| 493 |
-
[More Information Needed]
|
| 494 |
-
|
| 495 |
-
#### Who are the annotators?
|
| 496 |
-
|
| 497 |
-
[More Information Needed]
|
| 498 |
-
|
| 499 |
-
### Personal and Sensitive Information
|
| 500 |
-
|
| 501 |
-
[More Information Needed]
|
| 502 |
-
|
| 503 |
-
## Considerations for Using the Data
|
| 504 |
-
|
| 505 |
-
### Social Impact of Dataset
|
| 506 |
-
|
| 507 |
-
[More Information Needed]
|
| 508 |
-
|
| 509 |
-
### Discussion of Biases
|
| 510 |
-
|
| 511 |
-
[More Information Needed]
|
| 512 |
-
|
| 513 |
-
### Other Known Limitations
|
| 514 |
-
|
| 515 |
-
[More Information Needed]
|
| 516 |
|
| 517 |
## Additional Information
|
| 518 |
|
| 519 |
-
### Dataset Curators
|
| 520 |
-
|
| 521 |
-
[More Information Needed]
|
| 522 |
-
|
| 523 |
### Licensing Information
|
| 524 |
|
| 525 |
-
|
| 526 |
|
| 527 |
### Citation Information
|
| 528 |
|
| 529 |
-
```
|
| 530 |
@article{2019t5,
|
| 531 |
author = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
|
| 532 |
title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
|
|
|
|
| 135 |
paperswithcode_id: mc4
|
| 136 |
---
|
| 137 |
|
| 138 |
+
# Dataset Card for mC4-sampling
|
| 139 |
|
| 140 |
## Table of Contents
|
| 141 |
|
| 142 |
+
- [Dataset Card for mC4-sampling](#dataset-card-for-mc4-sampling)
|
| 143 |
- [Table of Contents](#table-of-contents)
|
| 144 |
- [Dataset Description](#dataset-description)
|
| 145 |
- [Dataset Summary](#dataset-summary)
|
| 146 |
+
- [Dataset Sampling](#dataset-sampling)
|
| 147 |
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
| 148 |
- [Languages](#languages)
|
| 149 |
- [Dataset Structure](#dataset-structure)
|
| 150 |
- [Data Instances](#data-instances)
|
| 151 |
- [Data Fields](#data-fields)
|
| 152 |
- [Data Splits](#data-splits)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
- [Additional Information](#additional-information)
|
| 154 |
- [Dataset Curators](#dataset-curators)
|
| 155 |
- [Licensing Information](#licensing-information)
|
|
|
|
| 158 |
|
| 159 |
## Dataset Description
|
| 160 |
|
| 161 |
+
- **Homepage:** https://huggingface.co/bertin-project/bertin-roberta-base-spanish
|
|
|
|
| 162 |
|
| 163 |
### Dataset Summary
|
| 164 |
|
| 165 |
+
This dataset builds upon the AllenAI version of the original [mC4](https://huggingface.co/datasets/allenai/c4) and adds sampling methods to perform perplexity-based filtering on the fly. Please, refer to [BERTIN Project](https://huggingface.co/bertin-project/bertin-roberta-base-spanish).
|
| 166 |
+
|
| 167 |
+
The original dataset is mC4, the multilingual colossal, cleaned version of Common Crawl's web crawl corpus. Based on Common Crawl dataset: "https://commoncrawl.org".
|
| 168 |
+
|
| 169 |
+
108 languages are available and are reported in the [`mc4` dataset](https://huggingface.co/datasets/mc4#dataset-summary).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 170 |
|
| 171 |
You can load the mC4 subset of any language like this:
|
| 172 |
|
|
|
|
| 184 |
mc4_subset_with_five_languages = load_dataset("mc4", languages=["en", "fr", "es", "de", "zh"])
|
| 185 |
```
|
| 186 |
|
| 187 |
+
### Dataset Sampling
|
| 188 |
+
|
| 189 |
+
There are 3 main different ways of getting sampled versions of mc4 using this dataset.
|
| 190 |
+
|
| 191 |
+
#### Random
|
| 192 |
+
|
| 193 |
+
Arguably, the simplest of methods. It keeps a document based on a probability threshold we called `factor`. It defaults to `0.5` for random sampling:
|
| 194 |
+
|
| 195 |
+
```python
|
| 196 |
+
def _should_keep_doc_random(self, doc, factor=None, **kwargs):
|
| 197 |
+
factor = 0.5 if factor is None else factor
|
| 198 |
+
return self.rng.uniform() <= factor
|
| 199 |
+
```
|
| 200 |
+
|
| 201 |
+
The way to use this sampling method is by adding an extra parameter to the instantiation of the dataset:
|
| 202 |
+
|
| 203 |
+
```python
|
| 204 |
+
from datasets import load_dataset
|
| 205 |
+
|
| 206 |
+
mc4random = load_dataset(
|
| 207 |
+
"bertin-project/mc4-sampling", "es",
|
| 208 |
+
split="train",
|
| 209 |
+
streaming=True,
|
| 210 |
+
sampling_method="random",
|
| 211 |
+
factor=0.5,
|
| 212 |
+
)
|
| 213 |
+
for sample in mc4random:
|
| 214 |
+
print(sample)
|
| 215 |
+
break
|
| 216 |
+
```
|
| 217 |
+
|
| 218 |
### Supported Tasks and Leaderboards
|
| 219 |
|
| 220 |
+
mC4-sampling is mainly intended to pretrain language models and word representations on a budget.
|
| 221 |
|
| 222 |
### Languages
|
| 223 |
|
|
|
|
| 260 |
|
| 261 |
### Data Splits
|
| 262 |
|
| 263 |
+
The same splits as in [mC4 are available](https://huggingface.co/datasets/mc4#data-splits).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 264 |
|
| 265 |
## Additional Information
|
| 266 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 267 |
### Licensing Information
|
| 268 |
|
| 269 |
+
BERTIN Project is releasing this dataset under the same terms AllenAI released mC4, that is, those of the ODC-BY. By using this, you are also bound by the Common Crawl terms of use in respect of the content contained in the dataset.
|
| 270 |
|
| 271 |
### Citation Information
|
| 272 |
|
| 273 |
+
```bibtex
|
| 274 |
@article{2019t5,
|
| 275 |
author = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
|
| 276 |
title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
|
mc4-sampling.py
CHANGED
|
@@ -340,7 +340,7 @@ class Mc4Sampling(datasets.GeneratorBasedBuilder):
|
|
| 340 |
probability = factor / quartile_range
|
| 341 |
return self.rng.uniform() < probability
|
| 342 |
|
| 343 |
-
def _should_keep_doc_gaussian(self, doc, factor=None, width=None, **kwargs):
|
| 344 |
perplexity = self.get_perplexity(doc)
|
| 345 |
width = (9 / 2) if width is None else width # width (spread) of the exponential curve
|
| 346 |
factor = 0.78 if factor is None else factor
|
|
|
|
| 340 |
probability = factor / quartile_range
|
| 341 |
return self.rng.uniform() < probability
|
| 342 |
|
| 343 |
+
def _should_keep_doc_gaussian(self, doc, factor=None, width=None, boundaries=None, **kwargs):
|
| 344 |
perplexity = self.get_perplexity(doc)
|
| 345 |
width = (9 / 2) if width is None else width # width (spread) of the exponential curve
|
| 346 |
factor = 0.78 if factor is None else factor
|