Hugging Face's logo

Datasets:
bigscience
/
xP3megds

Tasks:
Other
Languages:
Akan
Arabic
Assamese
Size:
100M<n<1B
ArXiv:
License:
Dataset card Files Community
1
Muennighoff commited on
Commit
693321c
•
1 Parent(s): 87eca22

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +348 -0
README.md ADDED
@@ -0,0 +1,348 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ annotations_creators:
3
+ - expert-generated
4
+ - crowdsourced
5
+ language:
6
+ - ak
7
+ - ar
8
+ - as
9
+ - bm
10
+ - bn
11
+ - ca
12
+ - code
13
+ - en
14
+ - es
15
+ - eu
16
+ - fon
17
+ - fr
18
+ - gu
19
+ - hi
20
+ - id
21
+ - ig
22
+ - ki
23
+ - kn
24
+ - lg
25
+ - ln
26
+ - ml
27
+ - mr
28
+ - ne
29
+ - nso
30
+ - ny
31
+ - or
32
+ - pa
33
+ - pt
34
+ - rn
35
+ - rw
36
+ - sn
37
+ - st
38
+ - sw
39
+ - ta
40
+ - te
41
+ - tn
42
+ - ts
43
+ - tum
44
+ - tw
45
+ - ur
46
+ - vi
47
+ - wo
48
+ - xh
49
+ - yo
50
+ - zh
51
+ - zu
52
+ programming_language:
53
+ - C
54
+ - C++
55
+ - C#
56
+ - Go
57
+ - Java
58
+ - JavaScript
59
+ - Lua
60
+ - PHP
61
+ - Python
62
+ - Ruby
63
+ - Rust
64
+ - Scala
65
+ - TypeScript
66
+ license:
67
+ - apache-2.0
68
+ multilinguality:
69
+ - multilingual
70
+ pretty_name: xP3
71
+ size_categories:
72
+ - 100M<n<1B
73
+ task_categories:
74
+ - other
75
+ ---
76
+
77
+ # Dataset Card for xP3
78
+
79
+ ## Table of Contents
80
+ - [Table of Contents](#table-of-contents)
81
+ - [Dataset Description](#dataset-description)
82
+ - [Dataset Summary](#dataset-summary)
83
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
84
+ - [Languages](#languages)
85
+ - [Dataset Structure](#dataset-structure)
86
+ - [Data Instances](#data-instances)
87
+ - [Data Fields](#data-fields)
88
+ - [Data Splits](#data-splits)
89
+ - [Dataset Creation](#dataset-creation)
90
+ - [Curation Rationale](#curation-rationale)
91
+ - [Source Data](#source-data)
92
+ - [Annotations](#annotations)
93
+ - [Additional Information](#additional-information)
94
+ - [Licensing Information](#licensing-information)
95
+ - [Citation Information](#citation-information)
96
+ - [Contributions](#contributions)
97
+
98
+ ## Dataset Description
99
+
100
+ - **Repository:** https://github.com/bigscience-workshop/xmtf
101
+ - **Paper:** [Crosslingual Generalization through Multitask Finetuning](https://arxiv.org/abs/2211.01786)
102
+ - **Point of Contact:** [Niklas Muennighoff](mailto:niklas@hf.co)
103
+
104
+ ### Dataset Summary
105
+
106
+ > xP3 (Crosslingual Public Pool of Prompts) is a collection of prompts & datasets across 46 of languages & 16 NLP tasks. It is used for the training of BLOOMZ and mT0, multilingual language models capable of following human instructions in dozens of languages zero-shot.
107
+
108
+ - **Creation:** The dataset can be recreated using instructions available [here](https://github.com/bigscience-workshop/xmtf#create-xp3). We provide this version to save processing time and ease reproducibility.
109
+ - **Languages:** 46 (Can be extended by [recreating with more splits](https://github.com/bigscience-workshop/xmtf#create-xp3))
110
+ - **xP3 Dataset Family:**
111
+
112
+ <table>
113
+ <tr>
114
+ <th>Name</th>
115
+ <th>Explanation</th>
116
+ <th>Example models</th>
117
+ </tr>
118
+ <tr>
119
+ <td><a href=https://huggingface.co/datasets/bigscience/xP3>xP3</a></t>
120
+ <td>Mixture of 13 training tasks in 46 languages with English prompts</td>
121
+ <td><a href=https://huggingface.co/bigscience/bloomz>bloomz</a> & <a href=https://huggingface.co/bigscience/mt0-xxl>mt0-xxl</a></td>
122
+ </tr>
123
+ <tr>
124
+ <td><a href=https://huggingface.co/datasets/bigscience/xP3mt>xP3mt</a></t>
125
+ <td>Mixture of 13 training tasks in 46 languages with prompts in 20 languages (machine-translated from English)</td>
126
+ <td><a href=https://huggingface.co/bigscience/bloomz-mt>bloomz-mt</a> & <a href=https://huggingface.co/bigscience/mt0-xxl-mt>mt0-xxl-mt</a></td>
127
+ </tr>
128
+ <tr>
129
+ <td><a href=https://huggingface.co/datasets/bigscience/xP3all>xP3all</a></t>
130
+ <td>xP3 + our evaluation datasets adding an additional 3 tasks for a total of 16 tasks in 46 languages with English prompts</td>
131
+ <td></td>
132
+ </tr>
133
+ <tr>
134
+ <td><a href=https://huggingface.co/datasets/bigscience/xP3megds>xP3megds</a></t>
135
+ <td><a href=https://github.com/bigscience-workshop/Megatron-DeepSpeed>Megatron-DeepSpeed</a> processed version of xP3</td>
136
+ <td><a href=https://huggingface.co/bigscience/bloomz>bloomz</a></td>
137
+ </tr>
138
+ <tr>
139
+ <td><a href=https://huggingface.co/datasets/Muennighoff/P3>P3</a></t>
140
+ <td>Repreprocessed version of the English-only <a href=https://huggingface.co/datasets/bigscience/P3>P3</a> with 8 training tasks</td>
141
+ <td><a href=https://huggingface.co/bigscience/bloomz-p3>bloomz-p3</a> & <a href=https://huggingface.co/bigscience/mt0-xxl-p3>mt0-xxl-p3</a></td>
142
+ </tr>
143
+ </table>
144
+
145
+ ## Dataset Structure
146
+
147
+ ### Data Instances
148
+
149
+ An example of "train" looks as follows:
150
+ ```json
151
+ {
152
+ "inputs": "Sentence 1: Fue académico en literatura metafísica, teología y ciencias clásicas.\nSentence 2: Fue académico en literatura metafísica, teología y ciencia clásica.\nQuestion: Can we rewrite Sentence 1 to Sentence 2? Yes or No?",
153
+ "targets": "Yes"
154
+ }
155
+ ```
156
+
157
+ ### Data Fields
158
+
159
+ The data fields are the same among all splits:
160
+ - `inputs`: the natural language input fed to the model
161
+ - `targets`: the natural language target that the model has to generate
162
+
163
+ ### Data Splits
164
+
165
+ The below table summarizes sizes per language (computed from the `merged_{lang}.jsonl` files). Due to languages like `tw` only being single sentence translation samples from Flores, their byte percentage is significantly lower than their sample percentage.
166
+
167
+ |Language|Kilobytes|%|Samples|%|
168
+ |--------|------:|-:|---:|-:|
169
+ |tw|106288|0.11|265071|0.34|
170
+ |bm|107056|0.11|265180|0.34|
171
+ |ak|108096|0.11|265071|0.34|
172
+ |eu|108112|0.11|269973|0.34|
173
+ |ca|110608|0.12|271191|0.34|
174
+ |fon|113072|0.12|265063|0.34|
175
+ |st|114080|0.12|265063|0.34|
176
+ |ki|115040|0.12|265180|0.34|
177
+ |tum|116032|0.12|265063|0.34|
178
+ |wo|122560|0.13|365063|0.46|
179
+ |ln|126304|0.13|365060|0.46|
180
+ |as|156256|0.16|265063|0.34|
181
+ |or|161472|0.17|265063|0.34|
182
+ |kn|165456|0.17|265063|0.34|
183
+ |ml|175040|0.18|265864|0.34|
184
+ |rn|192992|0.2|318189|0.4|
185
+ |nso|229712|0.24|915051|1.16|
186
+ |tn|235536|0.25|915054|1.16|
187
+ |lg|235936|0.25|915021|1.16|
188
+ |rw|249360|0.26|915043|1.16|
189
+ |ts|250256|0.26|915044|1.16|
190
+ |sn|252496|0.27|865056|1.1|
191
+ |xh|254672|0.27|915058|1.16|
192
+ |zu|263712|0.28|915061|1.16|
193
+ |ny|272128|0.29|915063|1.16|
194
+ |ig|325232|0.34|950097|1.2|
195
+ |yo|352784|0.37|918416|1.16|
196
+ |ne|393680|0.41|315754|0.4|
197
+ |pa|523248|0.55|339210|0.43|
198
+ |gu|560688|0.59|347499|0.44|
199
+ |sw|560896|0.59|1114455|1.41|
200
+ |mr|666240|0.7|417269|0.53|
201
+ |bn|832720|0.88|428843|0.54|
202
+ |ta|924496|0.97|410633|0.52|
203
+ |te|1332912|1.4|573364|0.73|
204
+ |ur|1918272|2.02|855756|1.08|
205
+ |vi|3101408|3.27|1667306|2.11|
206
+ |code|4330752|4.56|2707724|3.43|
207
+ |hi|4393696|4.63|1543441|1.96|
208
+ |zh|4589904|4.83|3560556|4.51|
209
+ |id|4606288|4.85|2627392|3.33|
210
+ |ar|4677264|4.93|2148955|2.72|
211
+ |fr|5546688|5.84|5055942|6.41|
212
+ |pt|6129584|6.46|3562772|4.52|
213
+ |es|7571808|7.98|5151349|6.53|
214
+ |en|37261104|39.25|31495184|39.93|
215
+ |total|94941936|100.0|78883588|100.0|
216
+
217
+ ## Dataset Creation
218
+
219
+ ### Source Data
220
+
221
+ #### Training datasets
222
+
223
+ - Code Miscellaneous
224
+ - [CodeComplex](https://huggingface.co/datasets/codeparrot/codecomplex)
225
+ - [Docstring Corpus](https://huggingface.co/datasets/teven/code_docstring_corpus)
226
+ - [GreatCode](https://huggingface.co/datasets/great_code)
227
+ - [State Changes](https://huggingface.co/datasets/Fraser/python-state-changes)
228
+ - Closed-book QA
229
+ - [Hotpot QA](https://huggingface.co/datasets/hotpot_qa)
230
+ - [Trivia QA](https://huggingface.co/datasets/trivia_qa)
231
+ - [Web Questions](https://huggingface.co/datasets/web_questions)
232
+ - [Wiki QA](https://huggingface.co/datasets/wiki_qa)
233
+ - Extractive QA
234
+ - [Adversarial QA](https://huggingface.co/datasets/adversarial_qa)
235
+ - [CMRC2018](https://huggingface.co/datasets/cmrc2018)
236
+ - [DRCD](https://huggingface.co/datasets/clue)
237
+ - [DuoRC](https://huggingface.co/datasets/duorc)
238
+ - [MLQA](https://huggingface.co/datasets/mlqa)
239
+ - [Quoref](https://huggingface.co/datasets/quoref)
240
+ - [ReCoRD](https://huggingface.co/datasets/super_glue)
241
+ - [ROPES](https://huggingface.co/datasets/ropes)
242
+ - [SQuAD v2](https://huggingface.co/datasets/squad_v2)
243
+ - [xQuAD](https://huggingface.co/datasets/xquad)
244
+ - TyDI QA
245
+ - [Primary](https://huggingface.co/datasets/khalidalt/tydiqa-primary)
246
+ - [Goldp](https://huggingface.co/datasets/khalidalt/tydiqa-goldp)
247
+ - Multiple-Choice QA
248
+ - [ARC](https://huggingface.co/datasets/ai2_arc)
249
+ - [C3](https://huggingface.co/datasets/c3)
250
+ - [CoS-E](https://huggingface.co/datasets/cos_e)
251
+ - [Cosmos](https://huggingface.co/datasets/cosmos)
252
+ - [DREAM](https://huggingface.co/datasets/dream)
253
+ - [MultiRC](https://huggingface.co/datasets/super_glue)
254
+ - [OpenBookQA](https://huggingface.co/datasets/openbookqa)
255
+ - [PiQA](https://huggingface.co/datasets/piqa)
256
+ - [QUAIL](https://huggingface.co/datasets/quail)
257
+ - [QuaRel](https://huggingface.co/datasets/quarel)
258
+ - [QuaRTz](https://huggingface.co/datasets/quartz)
259
+ - [QASC](https://huggingface.co/datasets/qasc)
260
+ - [RACE](https://huggingface.co/datasets/race)
261
+ - [SciQ](https://huggingface.co/datasets/sciq)
262
+ - [Social IQA](https://huggingface.co/datasets/social_i_qa)
263
+ - [Wiki Hop](https://huggingface.co/datasets/wiki_hop)
264
+ - [WiQA](https://huggingface.co/datasets/wiqa)
265
+ - Paraphrase Identification
266
+ - [MRPC](https://huggingface.co/datasets/super_glue)
267
+ - [PAWS](https://huggingface.co/datasets/paws)
268
+ - [PAWS-X](https://huggingface.co/datasets/paws-x)
269
+ - [QQP](https://huggingface.co/datasets/qqp)
270
+ - Program Synthesis
271
+ - [APPS](https://huggingface.co/datasets/codeparrot/apps)
272
+ - [CodeContests](https://huggingface.co/datasets/teven/code_contests)
273
+ - [JupyterCodePairs](https://huggingface.co/datasets/codeparrot/github-jupyter-text-code-pairs)
274
+ - [MBPP](https://huggingface.co/datasets/Muennighoff/mbpp)
275
+ - [NeuralCodeSearch](https://huggingface.co/datasets/neural_code_search)
276
+ - [XLCoST](https://huggingface.co/datasets/codeparrot/xlcost-text-to-code)
277
+ - Structure-to-text
278
+ - [Common Gen](https://huggingface.co/datasets/common_gen)
279
+ - [Wiki Bio](https://huggingface.co/datasets/wiki_bio)
280
+ - Sentiment
281
+ - [Amazon](https://huggingface.co/datasets/amazon_polarity)
282
+ - [App Reviews](https://huggingface.co/datasets/app_reviews)
283
+ - [IMDB](https://huggingface.co/datasets/imdb)
284
+ - [Rotten Tomatoes](https://huggingface.co/datasets/rotten_tomatoes)
285
+ - [Yelp](https://huggingface.co/datasets/yelp_review_full)
286
+ - Simplification
287
+ - [BiSECT](https://huggingface.co/datasets/GEM/BiSECT)
288
+ - Summarization
289
+ - [CNN Daily Mail](https://huggingface.co/datasets/cnn_dailymail)
290
+ - [Gigaword](https://huggingface.co/datasets/gigaword)
291
+ - [MultiNews](https://huggingface.co/datasets/multi_news)
292
+ - [SamSum](https://huggingface.co/datasets/samsum)
293
+ - [Wiki-Lingua](https://huggingface.co/datasets/GEM/wiki_lingua)
294
+ - [XLSum](https://huggingface.co/datasets/GEM/xlsum)
295
+ - [XSum](https://huggingface.co/datasets/xsum)
296
+ - Topic Classification
297
+ - [AG News](https://huggingface.co/datasets/ag_news)
298
+ - [DBPedia](https://huggingface.co/datasets/dbpedia_14)
299
+ - [TNEWS](https://huggingface.co/datasets/clue)
300
+ - [TREC](https://huggingface.co/datasets/trec)
301
+ - [CSL](https://huggingface.co/datasets/clue)
302
+ - Translation
303
+ - [Flores-200](https://huggingface.co/datasets/Muennighoff/flores200)
304
+ - [Tatoeba](https://huggingface.co/datasets/Helsinki-NLP/tatoeba_mt)
305
+ - Word Sense disambiguation
306
+ - [WiC](https://huggingface.co/datasets/super_glue)
307
+ - [XL-WiC](https://huggingface.co/datasets/pasinit/xlwic)
308
+
309
+ #### Evaluation datasets (included in [xP3all](https://huggingface.co/datasets/bigscience/xP3all) except for HumanEval)
310
+
311
+ - Natural Language Inference
312
+ - [ANLI](https://huggingface.co/datasets/anli)
313
+ - [CB](https://huggingface.co/datasets/super_glue)
314
+ - [RTE](https://huggingface.co/datasets/super_glue)
315
+ - [XNLI](https://huggingface.co/datasets/xnli)
316
+ - Coreference Resolution
317
+ - [Winogrande](https://huggingface.co/datasets/winogrande)
318
+ - [XWinograd](https://huggingface.co/datasets/Muennighoff/xwinograd)
319
+ - Program Synthesis
320
+ - [HumanEval](https://huggingface.co/datasets/openai_humaneval)
321
+ - Sentence Completion
322
+ - [COPA](https://huggingface.co/datasets/super_glue)
323
+ - [Story Cloze](https://huggingface.co/datasets/story_cloze)
324
+ - [XCOPA](https://huggingface.co/datasets/xcopa)
325
+ - [XStoryCloze](https://huggingface.co/datasets/Muennighoff/xstory_cloze)
326
+
327
+ ## Additional Information
328
+
329
+ ### Licensing Information
330
+
331
+ The dataset is released under Apache 2.0.
332
+
333
+ ### Citation Information
334
+
335
+ ```bibtex
336
+ @misc{muennighoff2022crosslingual,
337
+ title={Crosslingual Generalization through Multitask Finetuning},
338
+ author={Niklas Muennighoff and Thomas Wang and Lintang Sutawika and Adam Roberts and Stella Biderman and Teven Le Scao and M Saiful Bari and Sheng Shen and Zheng-Xin Yong and Hailey Schoelkopf and Xiangru Tang and Dragomir Radev and Alham Fikri Aji and Khalid Almubarak and Samuel Albanie and Zaid Alyafeai and Albert Webson and Edward Raff and Colin Raffel},
339
+ year={2022},
340
+ eprint={2211.01786},
341
+ archivePrefix={arXiv},
342
+ primaryClass={cs.CL}
343
+ }
344
+ ```
345
+
346
+ ### Contributions
347
+
348
+ Thanks to the contributors of [promptsource](https://github.com/bigscience-workshop/promptsource/graphs/contributors) for adding many prompts used in this dataset.