Shadowgraph
About
Sign in

bigscience/xP3

Dataset

113

stars

65

commits

2

linked in READMEs

May 30, 2023

updated

README

Dataset Card for xP3

Table of Contents

Dataset Description

Dataset Summary

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.

  • Creation: The dataset can be recreated using instructions available here. We provide this version to save processing time and ease reproducibility.
  • Languages: 46 (Can be extended by recreating with more splits)
  • xP3 Dataset Family:
NameExplanationExample models
xP3x Mixture of 17 tasks in 277 languages with English promptsWIP - Join us at Project Aya @C4AI to help!
xP3 Mixture of 13 training tasks in 46 languages with English promptsbloomz & mt0-xxl
xP3mt Mixture of 13 training tasks in 46 languages with prompts in 20 languages (machine-translated from English)bloomz-mt & mt0-xxl-mt
xP3all xP3 + evaluation datasets adding an additional 3 tasks for a total of 16 tasks in 46 languages with English prompts
xP3megds Megatron-DeepSpeed processed version of xP3bloomz
P3 Repreprocessed version of the English-only P3 with 8 training tasksbloomz-p3 & mt0-xxl-p3

Dataset Structure

Data Instances

An example of "train" looks as follows:

{
"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?",
"targets": "Yes" 
}

Data Fields

The data fields are the same among all splits:

  • inputs: the natural language input fed to the model
  • targets: the natural language target that the model has to generate

Data Splits

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. Adding a new language is very simple, you can take this script adding Russian as an example.

LanguageKilobytes%Samples%
tw1062880.112650710.34
bm1070560.112651800.34
ak1080960.112650710.34
eu1081120.112699730.34
ca1106080.122711910.34
fon1130720.122650630.34
st1140800.122650630.34
ki1150400.122651800.34
tum1160320.122650630.34
wo1225600.133650630.46
ln1263040.133650600.46
as1562560.162650630.34
or1614720.172650630.34
kn1654560.172650630.34
ml1750400.182658640.34
rn1929920.23181890.4
nso2297120.249150511.16
tn2355360.259150541.16
lg2359360.259150211.16
rw2493600.269150431.16
ts2502560.269150441.16
sn2524960.278650561.1
xh2546720.279150581.16
zu2637120.289150611.16
ny2721280.299150631.16
ig3252320.349500971.2
yo3527840.379184161.16
ne3936800.413157540.4
pa5232480.553392100.43
gu5606880.593474990.44
sw5608960.5911144551.41
mr6662400.74172690.53
bn8327200.884288430.54
ta9244960.974106330.52
te13329121.45733640.73
ur19182722.028557561.08
vi31014083.2716673062.11
code43307524.5627077243.43
hi43936964.6315434411.96
zh45899044.8335605564.51
id46062884.8526273923.33
ar46772644.9321489552.72
fr55466885.8450559426.41
pt61295846.4635627724.52
es75718087.9851513496.53
en3726110439.253149518439.93
total94941936100.078883588100.0

Dataset Creation

Source Data

Training datasets

Evaluation datasets (included in xP3all except for NLI datasets & HumanEval)

Additional Information

Licensing Information

The dataset is released under Apache 2.0.

Citation Information

@article{muennighoff2022crosslingual,
  title={Crosslingual generalization through multitask finetuning},
  author={Muennighoff, Niklas and Wang, Thomas and Sutawika, Lintang and Roberts, Adam and Biderman, Stella and Scao, Teven Le and Bari, M Saiful and Shen, Sheng and Yong, Zheng-Xin and Schoelkopf, Hailey and others},
  journal={arXiv preprint arXiv:2211.01786},
  year={2022}
}

Contributions

Thanks to the contributors of promptsource for adding many prompts used in this dataset.

Contributors

Muennighoff

65 commits

Similar repos

bigscience/xP3megds

Dataset

3

bigscience-workshop/xmtf

Crosslingual Generalization through Multitask Finetuning

534

bigscience/bloomz

Model

512

Muennighoff/xP3x

Dataset

95

bigscience/bloomz-3b

Model

81

bigscience/bloomz-1b1

Model

33

bigscience/bloomz-1b7

Model

23

bigscience/bloomz-7b1-mt

Model

144

Linked in READMEs

bigscience-workshop/promptsource

Toolkit for creating, sharing and using natural language prompts.

3,033

README badge

bigscience-workshop/xmtf

Crosslingual Generalization through Multitask Finetuning

534

README badge

bigscience/xP3

Dataset

113

stars

65

commits

2

linked in READMEs

May 30, 2023

updated

README

Dataset Card for xP3

Table of Contents

Dataset Description

Dataset Summary

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.

  • Creation: The dataset can be recreated using instructions available here. We provide this version to save processing time and ease reproducibility.
  • Languages: 46 (Can be extended by recreating with more splits)
  • xP3 Dataset Family:
NameExplanationExample models
xP3x Mixture of 17 tasks in 277 languages with English promptsWIP - Join us at Project Aya @C4AI to help!
xP3 Mixture of 13 training tasks in 46 languages with English promptsbloomz & mt0-xxl
xP3mt Mixture of 13 training tasks in 46 languages with prompts in 20 languages (machine-translated from English)bloomz-mt & mt0-xxl-mt
xP3all xP3 + evaluation datasets adding an additional 3 tasks for a total of 16 tasks in 46 languages with English prompts
xP3megds Megatron-DeepSpeed processed version of xP3bloomz
P3 Repreprocessed version of the English-only P3 with 8 training tasksbloomz-p3 & mt0-xxl-p3

Dataset Structure

Data Instances

An example of "train" looks as follows:

{
"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?",
"targets": "Yes" 
}

Data Fields

The data fields are the same among all splits:

  • inputs: the natural language input fed to the model
  • targets: the natural language target that the model has to generate

Data Splits

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. Adding a new language is very simple, you can take this script adding Russian as an example.

LanguageKilobytes%Samples%
tw1062880.112650710.34
bm1070560.112651800.34
ak1080960.112650710.34
eu1081120.112699730.34
ca1106080.122711910.34
fon1130720.122650630.34
st1140800.122650630.34
ki1150400.122651800.34
tum1160320.122650630.34
wo1225600.133650630.46
ln1263040.133650600.46
as1562560.162650630.34
or1614720.172650630.34
kn1654560.172650630.34
ml1750400.182658640.34
rn1929920.23181890.4
nso2297120.249150511.16
tn2355360.259150541.16
lg2359360.259150211.16
rw2493600.269150431.16
ts2502560.269150441.16
sn2524960.278650561.1
xh2546720.279150581.16
zu2637120.289150611.16
ny2721280.299150631.16
ig3252320.349500971.2
yo3527840.379184161.16
ne3936800.413157540.4
pa5232480.553392100.43
gu5606880.593474990.44
sw5608960.5911144551.41
mr6662400.74172690.53
bn8327200.884288430.54
ta9244960.974106330.52
te13329121.45733640.73
ur19182722.028557561.08
vi31014083.2716673062.11
code43307524.5627077243.43
hi43936964.6315434411.96
zh45899044.8335605564.51
id46062884.8526273923.33
ar46772644.9321489552.72
fr55466885.8450559426.41
pt61295846.4635627724.52
es75718087.9851513496.53
en3726110439.253149518439.93
total94941936100.078883588100.0

Dataset Creation

Source Data

Training datasets

Evaluation datasets (included in xP3all except for NLI datasets & HumanEval)

Additional Information

Licensing Information

The dataset is released under Apache 2.0.

Citation Information

@article{muennighoff2022crosslingual,
  title={Crosslingual generalization through multitask finetuning},
  author={Muennighoff, Niklas and Wang, Thomas and Sutawika, Lintang and Roberts, Adam and Biderman, Stella and Scao, Teven Le and Bari, M Saiful and Shen, Sheng and Yong, Zheng-Xin and Schoelkopf, Hailey and others},
  journal={arXiv preprint arXiv:2211.01786},
  year={2022}
}

Contributions

Thanks to the contributors of promptsource for adding many prompts used in this dataset.

Similar repos

bigscience/xP3megds

Dataset

3

bigscience-workshop/xmtf

Crosslingual Generalization through Multitask Finetuning

534

bigscience/bloomz

Model

512

Muennighoff/xP3x

Dataset

95

bigscience/bloomz-3b

Model

81

bigscience/bloomz-1b1

Model

33

bigscience/bloomz-1b7

Model

23

bigscience/bloomz-7b1-mt

Model

144

Linked in READMEs

bigscience-workshop/promptsource

Toolkit for creating, sharing and using natural language prompts.

3,033

README badge

bigscience-workshop/xmtf

Crosslingual Generalization through Multitask Finetuning

534

README badge

Contributors

Muennighoff

65 commits