doi
stringlengths 10
10
| chunk-id
stringlengths 1
4
| chunk
stringlengths 1
1.66k
| id
stringlengths 10
10
| title
stringlengths 19
148
| summary
stringlengths 345
1.92k
| source
stringlengths 31
31
| authors
list | categories
list | comment
stringlengths 4
284
⌀ | journal_ref
stringclasses 14
values | primary_category
stringclasses 16
values | published
stringlengths 8
8
| updated
stringlengths 8
8
| references
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1907.11692
|
49
|
2019. Xlnet: Generalized autoregressive pretraining for language understanding. arXiv preprint
arXiv:1906.08237 .
Yang You, Jing Li, Jonathan Hseu, Xiaodan Song,
James Demmel, and Cho-Jui Hsieh. 2019. Reducing bert pre-training time from 3 days to 76 minutes.
arXiv preprint arXiv:1904.00962 .
Rowan Zellers, Ari Holtzman, Hannah Rashkin,
Yonatan Bisk, Ali Farhadi, Franziska Roesner, and
Yejin Choi. 2019. Defending against neural fake
news. arXiv preprint arXiv:1905.12616 .
Yukun Zhu, Ryan Kiros, Richard Zemel, Ruslan
Salakhutdinov, Raquel Urtasun, Antonio Torralba,
and Sanja Fidler. 2015. Aligning books and movies:
Towards story-like visual explanations by watching movies and reading books. In arXiv preprint
arXiv:1506.06724 .
Appendix for “RoBERTa: A Robustly
Optimized BERT Pretraining Approach”
A Full results on GLUE
In Table 8we present the full set of development
|
1907.11692
|
RoBERTa: A Robustly Optimized BERT Pretraining Approach
|
Language model pretraining has led to significant performance gains but
careful comparison between different approaches is challenging. Training is
computationally expensive, often done on private datasets of different sizes,
and, as we will show, hyperparameter choices have significant impact on the
final results. We present a replication study of BERT pretraining (Devlin et
al., 2019) that carefully measures the impact of many key hyperparameters and
training data size. We find that BERT was significantly undertrained, and can
match or exceed the performance of every model published after it. Our best
model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results
highlight the importance of previously overlooked design choices, and raise
questions about the source of recently reported improvements. We release our
models and code.
|
http://arxiv.org/pdf/1907.11692
|
[
"Yinhan Liu",
"Myle Ott",
"Naman Goyal",
"Jingfei Du",
"Mandar Joshi",
"Danqi Chen",
"Omer Levy",
"Mike Lewis",
"Luke Zettlemoyer",
"Veselin Stoyanov"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20190726
|
20190726
|
[
{
"id": "1906.01604"
},
{
"id": "1904.09223"
},
{
"id": "1806.02847"
},
{
"id": "1704.04683"
},
{
"id": "1905.06290"
},
{
"id": "1506.06724"
},
{
"id": "1904.00962"
},
{
"id": "1606.08415"
},
{
"id": "1901.07291"
},
{
"id": "1907.11692"
},
{
"id": "1906.08237"
},
{
"id": "1801.06146"
},
{
"id": "1905.03197"
},
{
"id": "1907.10529"
},
{
"id": "1904.09482"
},
{
"id": "1905.12616"
},
{
"id": "1901.11504"
},
{
"id": "1903.07785"
}
] |
1907.11692
|
50
|
Optimized BERT Pretraining Approach”
A Full results on GLUE
In Table 8we present the full set of development
set results for RoBERTa. We present results for
aLARGE configuration that follows BERT LARGE ,
as well as a BASE configuration that follows
BERT BASE.B Pretraining Hyperparameters
Table 9describes the hyperparameters for pretraining of RoBERTa LARGE and RoBERTa BASE
C Finetuning Hyperparameters
Finetuning hyperparameters for RACE, SQuAD
and GLUE are given in Table 10. We select the
best hyperparameter values based on the median
of 5 random seeds for each task.
MNLI QNLI QQP RTE SST MRPC CoLA STS
RoBERTa BASE
+ all data + 500k steps 87.6 92.8 91.9 78.7 94.8 90.2 63.6 91.2
RoBERTa LARGE
with B OOKS + W IKI 89.0 93.9 91.9 84.5 95.3 90.2 66.3 91.6
+ additional data ( §3.2) 89.3 94.0 92.0 82.7 95.6 91.4 66.1 92.2
|
1907.11692
|
RoBERTa: A Robustly Optimized BERT Pretraining Approach
|
Language model pretraining has led to significant performance gains but
careful comparison between different approaches is challenging. Training is
computationally expensive, often done on private datasets of different sizes,
and, as we will show, hyperparameter choices have significant impact on the
final results. We present a replication study of BERT pretraining (Devlin et
al., 2019) that carefully measures the impact of many key hyperparameters and
training data size. We find that BERT was significantly undertrained, and can
match or exceed the performance of every model published after it. Our best
model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results
highlight the importance of previously overlooked design choices, and raise
questions about the source of recently reported improvements. We release our
models and code.
|
http://arxiv.org/pdf/1907.11692
|
[
"Yinhan Liu",
"Myle Ott",
"Naman Goyal",
"Jingfei Du",
"Mandar Joshi",
"Danqi Chen",
"Omer Levy",
"Mike Lewis",
"Luke Zettlemoyer",
"Veselin Stoyanov"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20190726
|
20190726
|
[
{
"id": "1906.01604"
},
{
"id": "1904.09223"
},
{
"id": "1806.02847"
},
{
"id": "1704.04683"
},
{
"id": "1905.06290"
},
{
"id": "1506.06724"
},
{
"id": "1904.00962"
},
{
"id": "1606.08415"
},
{
"id": "1901.07291"
},
{
"id": "1907.11692"
},
{
"id": "1906.08237"
},
{
"id": "1801.06146"
},
{
"id": "1905.03197"
},
{
"id": "1907.10529"
},
{
"id": "1904.09482"
},
{
"id": "1905.12616"
},
{
"id": "1901.11504"
},
{
"id": "1903.07785"
}
] |
1907.11692
|
51
|
+ additional data ( §3.2) 89.3 94.0 92.0 82.7 95.6 91.4 66.1 92.2
+ pretrain longer 300k 90.0 94.5 92.2 83.3 96.1 91.1 67.4 92.3
+ pretrain longer 500k 90.2 94.7 92.2 86.6 96.4 90.9 68.0 92.4
Table 8: Development set results on GLUE tasks for various co nfigurations of RoBERTa.
Hyperparam RoBERTa LARGE RoBERTa BASE
Number of Layers 24 12
Hidden size 1024 768
FFN inner hidden size 4096 3072
Attention heads 16 12
Attention head size 64 64
Dropout 0.1 0.1
Attention Dropout 0.1 0.1
Warmup Steps 30k 24k
Peak Learning Rate 4e-4 6e-4
Batch Size 8k 8k
Weight Decay 0.01 0.01
Max Steps 500k 500k
Learning Rate Decay Linear Linear
Adamǫ 1e-6 1e-6
Adamβ1 0.9 0.9
Adamβ2 0.98 0.98
Gradient Clipping 0.0 0.0
|
1907.11692
|
RoBERTa: A Robustly Optimized BERT Pretraining Approach
|
Language model pretraining has led to significant performance gains but
careful comparison between different approaches is challenging. Training is
computationally expensive, often done on private datasets of different sizes,
and, as we will show, hyperparameter choices have significant impact on the
final results. We present a replication study of BERT pretraining (Devlin et
al., 2019) that carefully measures the impact of many key hyperparameters and
training data size. We find that BERT was significantly undertrained, and can
match or exceed the performance of every model published after it. Our best
model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results
highlight the importance of previously overlooked design choices, and raise
questions about the source of recently reported improvements. We release our
models and code.
|
http://arxiv.org/pdf/1907.11692
|
[
"Yinhan Liu",
"Myle Ott",
"Naman Goyal",
"Jingfei Du",
"Mandar Joshi",
"Danqi Chen",
"Omer Levy",
"Mike Lewis",
"Luke Zettlemoyer",
"Veselin Stoyanov"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20190726
|
20190726
|
[
{
"id": "1906.01604"
},
{
"id": "1904.09223"
},
{
"id": "1806.02847"
},
{
"id": "1704.04683"
},
{
"id": "1905.06290"
},
{
"id": "1506.06724"
},
{
"id": "1904.00962"
},
{
"id": "1606.08415"
},
{
"id": "1901.07291"
},
{
"id": "1907.11692"
},
{
"id": "1906.08237"
},
{
"id": "1801.06146"
},
{
"id": "1905.03197"
},
{
"id": "1907.10529"
},
{
"id": "1904.09482"
},
{
"id": "1905.12616"
},
{
"id": "1901.11504"
},
{
"id": "1903.07785"
}
] |
1907.11692
|
52
|
Adamβ1 0.9 0.9
Adamβ2 0.98 0.98
Gradient Clipping 0.0 0.0
Table 9: Hyperparameters for pretraining RoBERTa LARGE and RoBERTa BASE.
Hyperparam RACE SQuAD GLUE
Learning Rate 1e-5 1.5e-5 {1e-5, 2e-5, 3e-5 }
Batch Size 16 48 {16, 32}
Weight Decay 0.1 0.01 0.1
Max Epochs 4 2 10
Learning Rate Decay Linear Linear Linear
Warmup ratio 0.06 0.06 0.06
Table 10: Hyperparameters for finetuning RoBERTa LARGE on RACE, SQuAD and GLUE.
|
1907.11692
|
RoBERTa: A Robustly Optimized BERT Pretraining Approach
|
Language model pretraining has led to significant performance gains but
careful comparison between different approaches is challenging. Training is
computationally expensive, often done on private datasets of different sizes,
and, as we will show, hyperparameter choices have significant impact on the
final results. We present a replication study of BERT pretraining (Devlin et
al., 2019) that carefully measures the impact of many key hyperparameters and
training data size. We find that BERT was significantly undertrained, and can
match or exceed the performance of every model published after it. Our best
model achieves state-of-the-art results on GLUE, RACE and SQuAD. These results
highlight the importance of previously overlooked design choices, and raise
questions about the source of recently reported improvements. We release our
models and code.
|
http://arxiv.org/pdf/1907.11692
|
[
"Yinhan Liu",
"Myle Ott",
"Naman Goyal",
"Jingfei Du",
"Mandar Joshi",
"Danqi Chen",
"Omer Levy",
"Mike Lewis",
"Luke Zettlemoyer",
"Veselin Stoyanov"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20190726
|
20190726
|
[
{
"id": "1906.01604"
},
{
"id": "1904.09223"
},
{
"id": "1806.02847"
},
{
"id": "1704.04683"
},
{
"id": "1905.06290"
},
{
"id": "1506.06724"
},
{
"id": "1904.00962"
},
{
"id": "1606.08415"
},
{
"id": "1901.07291"
},
{
"id": "1907.11692"
},
{
"id": "1906.08237"
},
{
"id": "1801.06146"
},
{
"id": "1905.03197"
},
{
"id": "1907.10529"
},
{
"id": "1904.09482"
},
{
"id": "1905.12616"
},
{
"id": "1901.11504"
},
{
"id": "1903.07785"
}
] |
2101.00027
|
0
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
Leo Gao Stella Biderman Sid Black Laurence Golding
Travis Hoppe Charles Foster Jason Phang Horace He
Anish Thite Noa Nabeshima Shawn Presser Connor Leahy
EleutherAI
[email protected]
Abstract
Recent work has demonstrated that increased
training dataset diversity improves general
cross-domain knowledge and downstream generalization capability for large-scale language
models. With this in mind, we present the
Pile: an 825 GiB English text corpus targeted at training large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets—both existing and newly
constructed—many of which derive from academic or professional sources. Our evaluation of the untuned performance of GPT-2 and
GPT-3 on the Pile shows that these models
struggle on many of its components, such as
academic writing. Conversely, models trained
on the Pile improve significantly over both
Raw CC and CC-100 on all components of the
Pile, while improving performance on downstream evaluations. Through an in-depth exploratory analysis, we document potentially
concerning aspects of the data for prospective
users. We make publicly available the code
used in its construction.1
1 Introduction
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
1
|
concerning aspects of the data for prospective
users. We make publicly available the code
used in its construction.1
1 Introduction
Recent breakthroughs in general-purpose language
modeling have demonstrated the effectiveness of
training massive models on large text corpora for
downstream applications (Radford et al., 2019;
Shoeybi et al., 2019; Raffel et al., 2019; Rosset,
2019; Brown et al., 2020; Lepikhin et al., 2020). As
the field continues to scale up language model training, the demand for high-quality massive text data
will continue to grow (Kaplan et al., 2020).
The growing need for data in language modeling
has caused most existing large-scale language models to turn to the Common Crawl for most or all of
their data (Brown et al., 2020; Raffel et al., 2019).
While training on the Common Crawl has been
effective, recent work has shown that dataset di1https://pile.eleuther.ai/versity leads to better downstream generalization
capability (Rosset, 2019). Additionally, large-scale
language models have been shown to effectively
acquire knowledge in a novel domain with only
relatively small amounts of training data from that
domain (Rosset, 2019; Brown et al., 2020; Carlini
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
2
|
acquire knowledge in a novel domain with only
relatively small amounts of training data from that
domain (Rosset, 2019; Brown et al., 2020; Carlini
et al., 2020). These results suggest that by mixing together a large number of smaller, high quality, diverse datasets, we can improve the general
cross-domain knowledge and downstream generalization capabilities of the model compared to models trained on only a handful of data sources.
To address this need, we introduce the Pile: a
825:18GiB English text dataset designed for training large scale language models. The Pile is composed of 22 diverse and high-quality datasets, including both established natural language processing datasets and several newly introduced ones.
In addition to its utility in training large language
models, the Pile can also serve as a broad-coverage
benchmark for cross-domain knowledge and generalization ability of language models.
We introduce new datasets derived from the following sources: PubMed Central, ArXiv, GitHub,
the FreeLaw Project, Stack Exchange, the US
Patent and Trademark Office, PubMed, Ubuntu
IRC, HackerNews, YouTube, PhilPapers, and NIH
ExPorter. We also introduce OpenWebText2 and
BookCorpus2, which are extensions of the original
OpenWebText (Gokaslan and Cohen, 2019) and
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
3
|
ExPorter. We also introduce OpenWebText2 and
BookCorpus2, which are extensions of the original
OpenWebText (Gokaslan and Cohen, 2019) and
BookCorpus (Zhu et al., 2015; Kobayashi, 2018)
datasets, respectively.
In addition, we incorporate several existing highquality datasets: Books3 (Presser, 2020), Project
Gutenberg (PG-19) (Rae et al., 2019), OpenSubtitles (Tiedemann, 2016), English Wikipedia,
DM Mathematics (Saxton et al., 2019), EuroParl
(Koehn, 2005), and the Enron Emails corpus (Klimt
and Yang, 2004). To supplement these, we also in1arXiv:2101.00027v1 [cs.CL] 31 Dec 2020
Figure 1: Treemap of Pile components by effective size.
troduce a new filtered subset of Common Crawl,
Pile-CC, with improved extraction quality.
Through our analyses, we confirm that the Pile is
significantly distinct from pure Common Crawl
data. Additionally, our evaluations show that the
existing GPT-2 and GPT-3 models perform poorly
on many components of the Pile, and that models
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
4
|
data. Additionally, our evaluations show that the
existing GPT-2 and GPT-3 models perform poorly
on many components of the Pile, and that models
trained on the Pile significantly outperform both
raw and filtered Common Crawl models. To complement the performance evaluations, we also perform an exploratory analysis of the text within the
Pile to provide a detailed picture of the data. We
hope that our extensive documentation of the construction and characteristics of the Pile will help
researchers make informed decisions about potential downstream applications.
Finally, we make publicly available the preprocessing code for the constituent datasets of the Pile and
the code for constructing alternative versions2. In
the interest of reproducibility, we also document
all processing performed on each dataset (and the
Pile as a whole) in as much detail as possible. For
further details about the processing of each dataset,
see Section 2 and Appendix C.
2https://github.com/EleutherAI/
the-pile1.1 Contributions
The core contributions of this paper are:
1.The introduction of a 825:18GiB englishlanguage dataset for language modeling combining 22 diverse sources.
2.The introduction of 14new language modeling datasets, which we expect to be of independent interest to researchers.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
5
|
2.The introduction of 14new language modeling datasets, which we expect to be of independent interest to researchers.
3.Evaluations demonstrating significant improvements across many domains by GPT-2sized models trained on this new dataset, compared to training on CC-100 and raw Common
Crawl.
4.The investigation and documentation of this
dataset, which we hope will better inform researchers about how to use it as well as motivate them to undertake similar investigations
of their own data.
2 The Pile Datasets
The Pile is composed of 22 constituent sub-datasets,
as shown in Table 1. Following Brown et al. (2020),
we increase the weights of higher quality components, with certain high-quality datasets such as
Wikipedia being seen up to 3 times (“epochs”) for
2
Component Raw Size Weight Epochs Effective Size Mean Document Size
Pile-CC 227.12 GiB 18.11% 1.0 227.12 GiB 4.33 KiB
PubMed Central 90.27 GiB 14.40% 2.0 180.55 GiB 30.55 KiB
Books3†100.96 GiB 12.07% 1.5 151.44 GiB 538.36 KiB
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
6
|
Books3†100.96 GiB 12.07% 1.5 151.44 GiB 538.36 KiB
OpenWebText2 62.77 GiB 10.01% 2.0 125.54 GiB 3.85 KiB
ArXiv 56.21 GiB 8.96% 2.0 112.42 GiB 46.61 KiB
Github 95.16 GiB 7.59% 1.0 95.16 GiB 5.25 KiB
FreeLaw 51.15 GiB 6.12% 1.5 76.73 GiB 15.06 KiB
Stack Exchange 32.20 GiB 5.13% 2.0 64.39 GiB 2.16 KiB
USPTO Backgrounds 22.90 GiB 3.65% 2.0 45.81 GiB 4.08 KiB
PubMed Abstracts 19.26 GiB 3.07% 2.0 38.53 GiB 1.30 KiB
Gutenberg (PG-19)†10.88 GiB 2.17% 2.5 27.19 GiB 398.73 KiB
OpenSubtitles†12.98 GiB 1.55% 1.5 19.47 GiB 30.48 KiB
Wikipedia (en)†6.38 GiB 1.53% 3.0 19.13 GiB 1.11 KiB
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
7
|
Wikipedia (en)†6.38 GiB 1.53% 3.0 19.13 GiB 1.11 KiB
DM Mathematics†7.75 GiB 1.24% 2.0 15.49 GiB 8.00 KiB
Ubuntu IRC 5.52 GiB 0.88% 2.0 11.03 GiB 545.48 KiB
BookCorpus2 6.30 GiB 0.75% 1.5 9.45 GiB 369.87 KiB
EuroParl†4.59 GiB 0.73% 2.0 9.17 GiB 68.87 KiB
HackerNews 3.90 GiB 0.62% 2.0 7.80 GiB 4.92 KiB
YoutubeSubtitles 3.73 GiB 0.60% 2.0 7.47 GiB 22.55 KiB
PhilPapers 2.38 GiB 0.38% 2.0 4.76 GiB 73.37 KiB
NIH ExPorter 1.89 GiB 0.30% 2.0 3.79 GiB 2.11 KiB
Enron Emails†0.88 GiB 0.14% 2.0 1.76 GiB 1.78 KiB
The Pile 825.18 GiB 1254.20 GiB 5.91 KiB
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
8
|
The Pile 825.18 GiB 1254.20 GiB 5.91 KiB
Table 1: Overview of datasets in the Pile before creating the held out sets. Raw Size is the size before any
up- or down-sampling. Weight is the percentage of bytes in the final dataset occupied by each dataset. Epochs
is the number of passes over each constituent dataset during a full epoch over the Pile. Effective Size is the
approximate number of bytes in the Pile occupied by each dataset. Datasets marked with a † are used with minimal
preprocessing from prior work.
each full epoch over the Pile. Detailed information
about the construction of each dataset is available
in Appendix C.
2.1 Pile-CC
Common Crawl is a collection of website crawls
from 2008 onwards, including raw web pages,
metadata and text extractions. Due to the raw nature of the dataset, Common Crawl has the advantage of including text from diverse domains,
but at the cost of varying quality data. Due to
this, use of Common Crawl typically necessitates well-designed extraction and filtering. Our
Common Crawl-based dataset, Pile-CC, uses jusText (Endrédy and Novák, 2013) on Web Archive
files (raw HTTP responses including page HTML)
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
9
|
files (raw HTTP responses including page HTML)
for extraction, which yields higher quality output
than directly using the WET files (extracted plaintext).2.2 PubMed Central
PubMed Central (PMC) is a subset of the PubMed
online repository for biomedical articles run by
the United States of America’s National Center
for Biotechnology Information (NCBI), providing
open, full-text access to nearly five million publications. Most publications indexed by PMC are
recent, and their inclusion is mandated for all NIH
funded research starting from 2008 by the NIH
Public Access Policy. We included PMC in the
hopes that it will benefit potential downstream applications to the medical domain.
2.3 Books3
Books3 is a dataset of books derived from a copy
of the contents of the Bibliotik private tracker
made available by Shawn Presser (Presser, 2020).
Bibliotik consists of a mix of fiction and nonfiction books and is almost an order of magnitude
3
larger than our next largest book dataset (BookCorpus2).We included Bibliotik because books are invaluable for long-range context modeling research
and coherent storytelling.
2.4 OpenWebText2
OpenWebText2 (OWT2) is a generalized web
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
10
|
and coherent storytelling.
2.4 OpenWebText2
OpenWebText2 (OWT2) is a generalized web
scrape dataset inspired by WebText (Radford et al.,
2019) and OpenWebTextCorpus (Gokaslan and Cohen, 2019). Similar to the original WebText, we
use net upvotes on Reddit submissions as a proxy
for outgoing link quality. OpenWebText2 includes
more recent content from Reddit submissions up
until 2020, content from multiple languages, document metadata, multiple dataset versions, and open
source replication code. We included OWT2 as a
high quality general purpose dataset.
2.5 ArXiv
ArXiv is a preprint server for research papers that
has operated since 1991. As shown in fig. 10, arXiv
papers are predominantly in the fields of Math,
Computer Science, and Physics. We included arXiv
in the hopes that it will be a source of high quality text and math knowledge, and benefit potential
downstream applications to research in these areas. ArXiv papers are written in LaTeX, a common
typesetting language for mathematics, computer
science, physics, and some adjacent fields. Training a language model to be able to generate papers
written in LaTeX could be a huge boon to the research community.
2.6 GitHub
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
11
|
written in LaTeX could be a huge boon to the research community.
2.6 GitHub
GitHub is a large corpus of open-source code repositories. Motivated by the ability of GPT-3 (Brown
et al., 2020) to generate plausible code completions
despite its training data not containing any explicitly gathered code datasets, we included GitHub in
the hopes that it would enable better downstream
performance on code-related tasks.
2.7 FreeLaw
The Free Law Project is a US-registered non-profit
that provides access to and analytical tools for academic studies in the legal realm. CourtListener,3
part of the Free Law Project, provides bulk downloads for millions of legal opinions from federal
and state courts. While the full dataset provides
multiple modalities of legal proceedings, including dockets, bibliographic information on judges,
3https://www.courtlistener.com/and other metadata, we focused specifically on
court opinions due to an abundance of full-text
entries. This data is entirely within the public domain.
2.8 Stack Exchange
The Stack Exchange Data Dump4contains an
anonymized set of all user-contributed content on
the Stack Exchange network, a popular collection
of websites centered around user-contributed questions and answers. It is one of the largest publicly
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
12
|
the Stack Exchange network, a popular collection
of websites centered around user-contributed questions and answers. It is one of the largest publicly
available repositories of question-answer pairs, and
covers a wide range of subjects—from programming, to gardening, to Buddhism. We included
Stack Exchange in the hopes that it will improve
the question answering capabilities of downstream
models on diverse domains.
2.9 USPTO Backgrounds
USPTO Backgrounds is a dataset of background
sections from patents granted by the United States
Patent and Trademark Office, derived from its published bulk archives5. A typical patent background
lays out the general context of the invention, gives
an overview of the technical field, and sets up the
framing of the problem space. We included USPTO
Backgrounds because it contains a large volume of
technical writing on applied subjects, aimed at a
non-technical audience.
2.10 Wikipedia (English)
Wikipedia is a standard source of high-quality text
for language modeling. In addition to being a
source of high quality, clean English text, it is also
valuable as it is written in expository prose, and
spans many domains.
2.11 PubMed Abstracts
PubMed Abstracts consists of the abstracts from 30
million publications in PubMed, the online repository for biomedical articles run by the National
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
13
|
spans many domains.
2.11 PubMed Abstracts
PubMed Abstracts consists of the abstracts from 30
million publications in PubMed, the online repository for biomedical articles run by the National
Library of Medicine. While the PMC (see Section
2.2) provides full-text access, the subset of coverage is significantly limited and biased towards recent publications. PubMed also incorporates MEDLINE, which expands the coverage of biomedical
abstracts from 1946 to present day.
4https://archive.org/details/
stackexchange
5https://bulkdata.uspto.gov/
4
2.12 Project Gutenberg
Project Gutenberg is a dataset of classic Western
literature. The specific Project Gutenberg derived
dataset we used, PG-19, consists of Project Gutenberg books from before 1919 (Rae et al., 2019),
which represent distinct styles from the more modern Books3 and BookCorpus. Additionally, the PG19 dataset is already being used for long-distance
context modeling.
2.13 OpenSubtitles
The OpenSubtitles dataset is an English language
dataset of subtitles from movies and television
shows gathered by Tiedemann (2016). Subtitles
provide an important source of natural dialog, as
well as an understanding of fictional formats other
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
14
|
shows gathered by Tiedemann (2016). Subtitles
provide an important source of natural dialog, as
well as an understanding of fictional formats other
than prose, which may prove useful for creative
writing generation tasks such as screenwriting,
speechwriting, and interactive storytelling.
2.14 DeepMind Mathematics
The DeepMind Mathematics dataset consists of a
collection of mathematical problems from topics
such as algebra, arithmetic, calculus, number theory, and probability, formatted as natural language
prompts (Saxton et al., 2019). One major weakness
of large language models has been performance
on mathematical tasks (Brown et al., 2020), which
may be due in part to a lack of math problems in
the training set. By explicitly including a dataset
of mathematical problems, we hope to improve the
mathematical ability of language models trained on
the Pile.
2.15 BookCorpus2
BookCorpus2 is an expanded version of the original BookCorpus (Zhu et al., 2015), a widely used
language modeling corpus consisting of books written by “as of yet unpublished authors.” BookCorpus is therefore unlikely to have significant overlap
with Project Gutenberg and Books3, which consist
of published books. BookCorpus is also commonly
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
15
|
with Project Gutenberg and Books3, which consist
of published books. BookCorpus is also commonly
used as dataset for training language models (Radford et al., 2018; Devlin et al., 2019; Liu et al.,
2019).
2.16 Ubuntu IRC
The Ubuntu IRC dataset is derived from the publicly available chatlogs6of all Ubuntu-related channels on the Freenode IRC chat server. Chatlog data
6https://irclogs.ubuntu.com/provides an opportunity to model real-time human
interactions, which feature a level of spontaneity
not typically found in other modes of social media.
2.17 EuroParl
EuroParl (Koehn, 2005) is a multilingual parallel
corpus originally introduced for machine translation but which has also seen use in several other
fields of NLP (Groves and Way, 2006; Van Halteren, 2008; Ciobanu et al., 2017). We use the most
current version at time of writing, which consists of
the proceedings of the European Parliament in 21
European languages from 1996 until 2012.
2.18 YouTube Subtitles
The YouTube Subtitles dataset is a parallel corpus of text gathered from human generated closedcaptions on YouTube. In addition to providing multilingual data, Youtube Subtitles is also a source of
educational content, popular culture, and natural
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
16
|
educational content, popular culture, and natural
dialog.
2.19 PhilPapers
The PhilPapers7dataset consists of open-access
philosophy publications from an international
database maintained by the Center for Digital Philosophy at the University of Western Ontario. We
included PhilPapers because it spans a wide body
of abstract, conceptual discourse, and its articles
contain high quality academic writing.
2.20 NIH Grant Abstracts:
ExPORTER
The NIH Grant abstracts provides a bulk-data repository for awarded applications through the ExPORTER8service covering the fiscal years 1985present. We included the dataset because it contains
examples of high-quality scientific writing.
2.21 Hacker News
Hacker News9is a link aggregator operated by Y
Combinator, a startup incubator and investment
fund. Users submit articles defined as “anything
that gratifies one’s intellectual curiosity,” but submitted articles tend to focus on topics in computer
science and entrepreneurship. Users can comment
on submitted stories, resulting in comment trees
discussing and critiquing submitted stories. We
7https://philpapers.org/
8https://exporter.nih.gov/
9https://news.ycombinator.com
5
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
17
|
7https://philpapers.org/
8https://exporter.nih.gov/
9https://news.ycombinator.com
5
scrape, parse, and include these comment trees
since we believe they provide high quality dialogue
and debate on niche topics.
2.22 Enron Emails
The Enron Emails dataset (Klimt and Yang, 2004)
is a valuable corpus commonly used for research
about the usage patterns of email. We included
Enron Emails to aid in understanding the modality
of email communications, which is typically not
found in any of our other datasets.
3 Benchmarking Language Models with
the Pile
While the Pile was conceived as a training dataset
for large-scale language models, its coverage of
multiple disparate domains makes it also suitable
as an evaluation dataset. In this section, we describe how the Pile can be used as a broad-coverage
dataset for benchmarking language models.
3.1 Benchmarking Guidelines
The Pile is provided as train, validation, and testing splits. The validation and testing components
each contain 0:1%of the data, sampled uniformly
at random. While this is a far smaller percentage
than most datasets, the sheer size of the dataset
results in over 1 GiB of validation and testing data
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
18
|
at random. While this is a far smaller percentage
than most datasets, the sheer size of the dataset
results in over 1 GiB of validation and testing data
each. We highlight that while we have made efforts to deduplicate documents within the Pile (See:
Section D.2), it is still possible that some documents are duplicated across the train/validation/test
splits.
Our preferred metric is bits per UTF-8 encoded
byte ( BPB). Bits per byte is preferred over bits per
character or perplexity when using Pile as a metric due to its invariance to different tokenization
schemes and the ambiguity of measuring characters in Unicode. To compute bits per byte from a
given negative log likelihood loss `, we compute
BPB = (LT=LB) log2(e`) = (LT=LB)`=ln(2) ,
whereLTis the length of the dataset in tokens and
LBis the length of the dataset in UTF-8 encoded
bytes. We find that LT=LBis0:29335 GPT-2tokens/byte across the Pile; dataset-specific values
ofLT=LBcan be found in Table 7.
3.2 Test Perplexity with GPT-2 and
GPT-3
We compute the test perplexity of the constituent
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
19
|
ofLT=LBcan be found in Table 7.
3.2 Test Perplexity with GPT-2 and
GPT-3
We compute the test perplexity of the constituent
datasets of the Pile using GPT-2 (Radford et al.,2019) and GPT-3 (Brown et al., 2020), shown in
Figure 2. We use all available versions of GPT-2,
and all four versions of GPT-3 available via the
OpenAI API. Because of the cost associated with
using the OpenAI API, we evaluate on one-tenth of
the respective test sets for most of the constituent
datasets. We report the perplexity converted to bits
per UTF-8 encoded byte ( BPB). Importantly, we
compute perplexity by evaluating each document
independently within each dataset, as opposed to
concatenating all documents as is common practice
for computing perplexity on large corpora.
Full details of the perplexity computation can be
found in Appendix E.2.
Unsurprisingly, larger language models generally
attain lower perplexity compared to smaller models.
Recent work has shown an increased focus on the
empirical scaling laws of language models (Kaplan
et al., 2020; Henighan et al., 2020). As such, we
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
20
|
Recent work has shown an increased focus on the
empirical scaling laws of language models (Kaplan
et al., 2020; Henighan et al., 2020). As such, we
investigate the scaling law for the GPT-2 and GPT3 families of models on perplexity evaluation on
the Pile. The scaling law relation for the GPT-3
family of models is shown in Figure 2.10The line
of best fit shown in the figure has a coefficient of
-0.1674 and an intercept of 2.5516.
Figure 2: Scaling law for performance of GPT-2/3 models. ‘Zero-shot’ refers to the fact that none of the models have been fine-tuned on data from the Pile.
Interestingly, while GPT-2 and GPT-3 were not
trained on the Pile, there still appears to be a clear
scaling law without diminishing returns. We hypothesize that this is due to the inherent generalization capability of these models. We leave a more
10While the sizes of GPT-3 models on the OpenAI API have
not been publicized, we assume here that ada,babbage ,
curie anddavinci models correspond to 2.7B, 6.7B, 13B
and 175B parameter models respectively.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
21
|
curie anddavinci models correspond to 2.7B, 6.7B, 13B
and 175B parameter models respectively.
6
rigorous analysis of zero-shot scaling laws to future
work.
3.3 Relative Componentwise GPT-3 Pile
Performance
Determining which components GPT-3 underperforms on provides information about which Pile
components are most dissimilar to the distribution
of text (web pages and books) that GPT-3 was
trained on. These components would thus make especially good candidates for supplementing GPT-3
training data. These results are also valuable for
determining which types of datasets to emphasize
for future iterations of the Pile.
Due to the difference in entropy of different
datasets, directly comparing perplexity of GPT-3
on different Pile components is not an accurate indication of relative performance. Ideally we would
train a GPT-3 model from scratch on the Pile and
compare the difference in loss per dataset with
that of the original GPT-3. Because of resource
constraints, we instead use a GPT-2 model trained
from scratch on the Pile (see Section 4) to construct a proxy measure. To construct our proxy,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
22
|
constraints, we instead use a GPT-2 model trained
from scratch on the Pile (see Section 4) to construct a proxy measure. To construct our proxy,
we first measure the improvement from the GPT2-Pile model to GPT-3 on each component. Then,
we normalize our results by setting the change on
OpenWebText2 to be zero. This computation is
shown in the equation below:
set=
LGPT3
set LGPT3
owt2
LGPT2Pile
set LGPT2Pile
owt2
Since GPT2-Pile was trained on both OWT2 and
the dataset we are evaluating, we expect the second
term in setto reflect the difference in the intrinsic
difficulty of the two datasets. Thus the total value
ofsetreflects how much harder the dataset we are
evaluating was for GPT-3 than OWT2, minus the
relative difficulty of the two tasks. As GPT-3 was
trained on data very similar to OWT2, this gives us
a proxy for how much better GPT-3 would do if it
were trained on the Pile.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
23
|
trained on data very similar to OWT2, this gives us
a proxy for how much better GPT-3 would do if it
were trained on the Pile.
The results are shown in Figure 3. As a sanity check, we observe that datasets that are contained in, or are extremely similar to, GPT-3’s
training set (Books3, Wikipedia (en), Pile-CC and
Project Gutenberg) score close to zero on our metric.GPT-3 appears to perform poorly on datasets
pertaining to research or academic writing like
PubMed Central, PubMed Abstracts, and ArXiv;
domain-specific datasets like FreeLaw, HackerNews, and USPTO Backgrounds; and on datasets
containing predominantly text distinct from natural language, like GitHub and DM Mathematics.
In addition, the majority of datasets see less of an
improvement than OpenWebText2. As such, we expect a GPT-3 sized model trained on Pile to perform
significantly better on research related tasks, software tasks, and symbol manipulation tasks than the
base model. Additionally, this experiment provides
evidence that the majority of Pile components are
not redundant with the predominantly web-based
GPT-3 training data.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
24
|
base model. Additionally, this experiment provides
evidence that the majority of Pile components are
not redundant with the predominantly web-based
GPT-3 training data.
We note that this metric is only a proxy for similarity, and that it could be confounded by dataset specific scaling effects. Although our results largely
accord with expectations, there are some puzzling
results, like the datasets on which GPT-3 outperformed GPT-2 Pile. We hypothesize that GPT-3
learns to be so good at these datasets that training on them explicitly does not notably benefit the
model’s performance. We leave a more rigorous
analysis of these effects for future work.
4 Evaluation
To confirm the effectiveness of the Pile for improving language modeling quality, we train
architecturally-identical 1.3 billion parameter models based on those in Brown et al. (2020) on different datasets and evaluate on the WikiText and
LAMBADA tasks as benchmarks of language modeling ability. We also report results on the Pile
as a measure of more cross-domain generalization.
4.1 Methodology
To ensure a fair comparison across datasets of different sizes, we decontaminate any instances of the
evaluation sets using the same 13-gram overlap filtering as in Brown et al. (2020) and downsample
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
25
|
evaluation sets using the same 13-gram overlap filtering as in Brown et al. (2020) and downsample
to 40GB to control for dataset size. As we control
for dataset size, we emphasize that our evaluation
is generous to CC-100 (en), which is about 1/3 the
size of the Pile in reality.
We compare the following datasets: the Pile, the En7
Component GPT-2 GPT-3
small medium large xl ada babbage curie davinci
Pile-CC 1.0878 0.9992 0.9582 0.9355 0.9212 0.8483 0.7849 0.7070
PubMed Central 1.0759 0.9788 0.9334 0.9044 0.8633 0.7792 0.7150 0.6544
Books3 1.1959 1.1063 1.0588 1.0287 0.9778 0.9005 0.8284 0.7052
OpenWebText2 1.1111 1.0073 0.9539 0.9171 0.8727 0.7921 0.7199 0.6242
ArXiv 1.3548 1.2305 1.1778 1.1381 1.0304 0.9259 0.8453 0.7702
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
26
|
ArXiv 1.3548 1.2305 1.1778 1.1381 1.0304 0.9259 0.8453 0.7702
Github 1.7912 1.3180 1.7909 1.6486 0.8761 0.7335 0.6415 0.5635
FreeLaw 1.0512 0.9321 0.9017 0.8747 0.8226 0.7381 0.6667 0.6006
Stack Exchange 1.2981 1.1075 1.0806 1.0504 1.0096 0.8839 0.8004 0.7321
USPTO Backgrounds 0.8288 0.7564 0.7202 0.6969 0.6799 0.6230 0.5752 0.5280
PubMed Abstracts 0.9524 0.8579 0.8108 0.7810 0.8130 0.7382 0.6773 0.6201
Gutenberg (PG-19) 1.2655 1.1140 1.0820 1.0829 0.9776 0.8749 0.7930 0.7115
OpenSubtitles 1.2465 1.1657 1.1324 1.1129 1.1116 1.0488 0.9875 0.9130
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
27
|
OpenSubtitles 1.2465 1.1657 1.1324 1.1129 1.1116 1.0488 0.9875 0.9130
Wikipedia (en) 1.1285 1.0213 0.9795 0.9655 0.8757 0.7863 0.7047 0.5953
DM Mathematics 2.6911 2.5448 2.4833 2.4377 2.3249 2.2015 2.1067 2.0228
Ubuntu IRC 1.8466 1.7187 1.6427 1.6024 1.3139 1.1968 1.0995 0.9915
BookCorpus2 1.1295 1.0498 1.0061 0.9783 0.9754 0.9041 0.8435 0.7788
EuroParl 2.3177 2.0204 1.8770 1.7650 1.0475 0.9363 0.8415 0.7519
HackerNews 1.4433 1.2794 1.3143 1.3361 1.1736 1.0875 1.0175 0.9457
YoutubeSubtitles 2.0387 1.8412 1.7355 1.6694 1.3407 1.1876 1.0639 0.9469
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
28
|
YoutubeSubtitles 2.0387 1.8412 1.7355 1.6694 1.3407 1.1876 1.0639 0.9469
PhilPapers 1.3203 1.2163 1.1688 1.1327 1.0362 0.9530 0.8802 0.8059
NIH ExPorter 0.9099 0.8323 0.7946 0.7694 0.7974 0.7326 0.6784 0.6239
Enron Emails 1.5888 1.4119 1.4535 1.4222 1.2634 1.1685 1.0990 1.0201
The Pile 1.2253 1.0928 1.0828 1.0468 0.9631 0.8718 0.7980 0.7177
Table 2: Test perplexity of the Pile using GPT-2 and GPT-3, converted to bits per UTF-8 encoded byte ( BPB).
Evaluation is performed on one-tenth of the test data of the Pile, on a per-document basis. Bold indicates the
best-performing model in each row.
glish component of the CC-100 dataset11(Wenzek
et al., 2019; Conneau et al., 2020), and a sample of
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
29
|
best-performing model in each row.
glish component of the CC-100 dataset11(Wenzek
et al., 2019; Conneau et al., 2020), and a sample of
raw CC WET files filtered for English-only.
4.2 Results
On traditional language modeling benchmarks, the
Pile improves significantly on WikiText and shows
negligible changes in LAMBADA. However, models trained on Pile improve significantly over both
Raw CC and CC-100 on all components of the
Pile, as shown in Table 4. This indicates that models trained on the Pile have greater cross-domain
generalization capabilities without compromising
performance on traditional benchmarks.
The magnitude of improvement over CC-100 per
set is shown in Figure 4. Unsurprisingly, there
is almost no improvement on Pile-CC. However,
the model trained on the Pile performs significantly better than either of the other models on
academic datasets such as ArXiv, Pubmed Central,
FreeLaw, and PhilPapers. It also improves signifi11The data was obtained from http://data.statmt.
org/cc-100/ .cantly on programming-related datasets like Github
and StackExchange, on EuroParl, due to the lack of
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
30
|
org/cc-100/ .cantly on programming-related datasets like Github
and StackExchange, on EuroParl, due to the lack of
multilingual text in either other dataset, and on DM
Mathematics, indicating a significant improvement
in mathematical ability.
Surprisingly, raw Common Crawl performs better
on the Pile BPB than CC-100, despite losing by a
significant margin on LAMBADA and WikiText.
We hypothesize that this is due to the perplexity
based filtering used in CC-100, where a language
model is trained on Wikipedia and all data with a
perplexity too high or too low is discarded. This
effectively discards any data too similar to or too
different from Wikipedia, which severely limits
the diversity of the collected data. This result
suggests that future work using Common Crawl
should take caution with filtering to preserve its
diversity.
5 Structural Statistics
In this section, we cover the Structural Statistics
of the dataset, which provide more coarse-grained
and statistical information about the Pile. In Sec8
Figure 3: Change in BPB from GPT-2 trained on Pile to GPT-3 zero-shot, relative to OpenWebText2 BPB change.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
31
|
Figure 3: Change in BPB from GPT-2 trained on Pile to GPT-3 zero-shot, relative to OpenWebText2 BPB change.
Dotted line indicates overall Pile change. Lower indicates better relative performance by GPT-3.
Dataset Size Pile (val) Pile (test) WikiText LAMBADA LAMBADA
(BPB) ( BPB) ( PPL) ( PPL) ( ACC)
The Pile 825 GiB 0.9281 0.9433 5.59 12.78 50.1
CC-100 (en) 300 GiB 1.3143 1.3293 8.27 11.78 49.7
Raw CC 45927 GiBy1.1180 1.1275 11.75 19.84 43.8
Table 3: Size-controlled evaluation results. Each dataset is deduplicated against all evaluation metrics and subsampled to approximately 40GB to control for the effects of dataset size. For LAMBADA, we use the variant of the
data introduced in Radford et al. (2019) and only evaluate the perplexity on the final token rather than the final
word. For WikiText, we report the perplexity per GPT-2 token. † indicates that the size is an estimate.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
32
|
word. For WikiText, we report the perplexity per GPT-2 token. † indicates that the size is an estimate.
tion 6, we provide a closer investigation and documentation of the textual content within the Pile
datasets.
5.1 Document Lengths and
Tokenization
Each dataset consists of a large number of documents. We analyze the distribution of document
lengths, as well as the number of bytes-per-token
using the GPT-2 tokenizer in order to put our ablations in context.
While the majority of documents in the Pile are
short, there is a long tail of very long documents
(Figure 5).
Since the GPT-2 BPE tokenizer is trained on WebText, the mean bytes per token is also a very rough
indicator of how syntactically different each Pile
component is from WebText. For instance, datasets
like NIH ExPorter, OpenWebText2 and Books3consist largely of ordinary text in a similar distribution to WebText, which is reflected in a greater
number of bytes per token. On the other hand,
many of the sets with the lowest bytes per token
are those which consist in large part of non-text
content (Github, ArXiv, Stack Exchange, and DM
Mathematics) or languages other than English (EuroParl).
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
33
|
content (Github, ArXiv, Stack Exchange, and DM
Mathematics) or languages other than English (EuroParl).
5.2 Language and Dialects
While only 13% of the world’s population speaks
English, the vast majority of NLP research is done
on English. For the Pile, we took a similar approach to the dataset used by Brown et al. (2020)
and focused predominantly on English, while also
not explicitly filtering out other languages when
collecting our own data. When evaluating a multilingual dataset, our main criteria for inclusion was
whether the English component of the dataset merited inclusion alone. We plan to create a fully multi9
Figure 4: Magnitude of BPB improvement of Pile model over CC-100 model on each test set.
Dataset The Pile CC-100 (en) Raw CC (en)
Pile-CC 0.9989 1.0873 1.0287
PubMed Central 0.6332 1.1311 0.9120
Books3 1.0734 1.2264 1.1366
OpenWebText2 0.9938 1.2222 1.0732
ArXiv 0.7945 1.8159 1.2642
Github 0.5597 1.6509 0.9301
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
34
|
ArXiv 0.7945 1.8159 1.2642
Github 0.5597 1.6509 0.9301
FreeLaw 0.6978 1.0221 0.9468
Stack Exchange 0.8152 1.5414 1.1292
USPTO Backgrounds 0.6731 0.8772 0.8455
PubMed Abstracts 0.7313 1.0193 0.9718
Gutenberg (PG-19) 1.1426 1.2780 1.2235
OpenSubtitles 1.0909 1.1827 1.2139
Wikipedia (en) 0.8961 1.1807 1.0252
DM Mathematics 1.5206 3.1774 2.6229
Ubuntu IRC 1.4085 2.1243 1.5691
BookCorpus2 1.0613 1.1346 1.0914
EuroParl 1.1202 2.7141 1.4917
HackerNews 1.0968 1.4352 1.2305
YoutubeSubtitles 1.4269 2.3287 1.5607
PhilPapers 1.1256 1.4269 1.2090
NIH ExPorter 0.7347 0.9713 0.9225
Enron Emails 0.8301 1.3300 1.0483
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
35
|
NIH ExPorter 0.7347 0.9713 0.9225
Enron Emails 0.8301 1.3300 1.0483
Table 4: Breakdown of BPB on Pile heldout test set.
Columns indicate the dataset each model is trained on;
rows indicate the evaluation dataset. Bold indicates the
best performing model in each row.
lingual expansion of the Pile as future work.
Using fasttext (Suárez et al., 2019a), we determine that the Pile is 97.4% English. We note that
due to issues with language identification, particularly with rare languages Caswell et al. (2020),
this methodology provides only a rough estimate
for English content and no reliable conclusions for
low-resource languages can be drawn.
Figure 5: Distribution of document lengths in Pile. The
highest 1 percentile of document length are considered
to be outliers and excluded from this plot.
6 Investigating and Documenting the
Datasets
As the scale of machine learning research has
grown, scrutiny has been placed on the ever
larger datasets that models are trained on (Prabhu
and Birhane, 2020; Biderman and Scheirer,
2020)
While this issue has been raised within AI ethics
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
36
|
and Birhane, 2020; Biderman and Scheirer,
2020)
While this issue has been raised within AI ethics
and bias research (Hovy and Spruit, 2016; Hutchinson et al., 2020; Blodgett et al., 2020), it has not
been a focal point of concern within the language
modeling community. Despite the proliferation
of work exploring and documenting issues with
datasets (Gebru et al., 2018; Bender and Friedman,
10
Figure 6: Mean bytes per GPT-2-token for each dataset
in the Pile. Error bars indicate standard deviation.
2018; Jo and Gebru, 2020), no dataset intended
to train massive language models has been seriously documented by its creators12. Therefore, our
analyses serve two goals: to address ethical concerns about the Pile, and to promote and normalize
the practice of engaging with the AI ethics literature.
Natural language processing technologies are
widely applicable and can be used in extremely
different contexts. What is and is not appropriate
data to train on can therefore vary wildly with the
application context. In our view, the best approach
is todocument rather than eliminate potentially concerning aspects of datasets13, particularly since the
purpose of the Pile is to train general-purpose language models. The primary goal of our documentation, therefore, is to empower NLP researchers to
make informed decisions.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
37
|
purpose of the Pile is to train general-purpose language models. The primary goal of our documentation, therefore, is to empower NLP researchers to
make informed decisions.
6.1 Documenting Methods
To document the Pile, we chose to implement two
frameworks that have been proposed by methodologists and ethics researchers. The first, the
datasheets methodology (Gebru et al., 2018), is a
general purpose methodology that is recommended
by several methodologists (Raji and Yang, 2019;
Biderman and Scheirer, 2020) and appears to be
used more frequently by practitioners than alterna12Brown et al. (2020) discusses ethical issues surrounding
their model , but do not discuss those surrounding the training
dataset itself.
13That said, we did exclude several datasets, see Appendix
B for details.tives (Seck et al., 2018; Costa-jussà et al., 2020;
Thieme et al., 2020). The second, the data statements methodology (Bender and Friedman, 2018),
was proposed specifically for natural language processing and has been well received by the NLP
community. Our datasheet and data statement will
be featured in the GitHub repository where the
code for the Pile is stored and will also be available
as separate documents on arXiv (Biderman et al.,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
38
|
be featured in the GitHub repository where the
code for the Pile is stored and will also be available
as separate documents on arXiv (Biderman et al.,
2021; Biderman, 2021).
In addition to the datasheet and data statement,
there is additional information that may be helpful
to people training language models that these documents do not cover. In the rest of this section we
investigate and document in greater detail some of
this additional contextual information.
6.2 Topical Distribution
In order to better understand the specific subject
matter covered by the Pile, we performed a topic
modeling analysis on its components. Using Gensim (Rehurek et al., 2011), we trained 16-topic Latent Dirichlet Allocation (Blei et al., 2003) models
on each component of the validation set of the Pile
concurrently, in an online fashion (Hoffman et al.,
2010). We filtered the Pile for English only for this
analysis. Afterwards, we computed the perplexity of the Common Crawl-derived (Pile-CC) topic
model on the document sets of the other components. In this way, we provide a rough measure of
the degree to which parts of the Pile contain topics
not well covered within Common Crawl.
In Figure 7, these cross-component perplexities are
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
39
|
the degree to which parts of the Pile contain topics
not well covered within Common Crawl.
In Figure 7, these cross-component perplexities are
shown, with a vertical line indicating the perplexity
of the Pile-CC topic model evaluated on the documents of OpenWebText2. This component was
chosen as a baseline of comparison for similar reasons as in the previous evaluation: it is derived in a
similar manner (filtered crawls of the open web) as
the Common Crawl, and thus is expected to contain
a similar distribution of topics. Although Pile-CC
is somewhat diverse in its content, several of the
Pile’s other components deviate from it strongly in
their topical focus, as evidenced by higher perplexity on Github, PhilPapers, and EuroParl.
We also documented the topical clusters inferred
from our LDA models for each component, which
we provide in Appendix C. As expected, though
the larger CC-derived component itself represents a
diversity of content—including politics, education,
11
sports and entertainment—the content clusters it
misses become apparent when compared qualitatively to other components of the Pile. Notably, the
data modes covering programming, logic, physics,
and legal knowledge appear largely absent.
6.3 Pejorative Content
Due to the wide diversity in origins, it is possible
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
40
|
data modes covering programming, logic, physics,
and legal knowledge appear largely absent.
6.3 Pejorative Content
Due to the wide diversity in origins, it is possible
for the Pile to contain pejorative, sexually explicit,
or otherwise objectionable content. As this content
may not be desirable for some use cases, we break
down profanity on a per-dataset level.
We used the profanity-checker Python
package (Zhou, 2019). This package includes a
“toxicity model” trained on multiple profanity lists
as well as the Wikidetox Toxic Comment Dataset
(Wulczyn et al., 2016) and classifies a given string
as being profane or not profane.
We considered only the English sentences in
each dataset using the same language classifier from Section 3.7. We did this since
profanity-checker is built for English and
other languages may improperly impact the results.
For instance, the German nominative/accusative
feminine/plural definite article "die" is flagged as
being profane regardless of context. We split each
sentence into words and computed the percentage
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
41
|
being profane regardless of context. We split each
sentence into words and computed the percentage
of words that are flagged as profane for each component of the Pile. We emphasize that this methodology is only a proxy for profanity, given the complexity of determining whether a given word or
phrase is profane in context.
As shown in Figure 8, the Pile as a whole appears
less profane than Pile-CC. Further, the majority of
Pile components appear less profane than Pile-CC
as well.
We also broke each dataset down on a sentence
level, to allow profanity-checker to check
entire sentences. Splitting datasets by sentence
allows for additional context to be considered when
determining whether content is pejorative. Our
results are shown in Figure 12.
6.4 Bias and Sentiment Co-occurrence
As language models may pick up unexpected biases
from the training data, we performed a preliminary
analysis of the different components that make up
the Pile. Because models with different characteristics may be trained on the Pile, we aimed to
document the biases of the data and not a specificmodel. We primarily focus on co-occurrence tests,
where we analyzed what words occur in the same
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
42
|
document the biases of the data and not a specificmodel. We primarily focus on co-occurrence tests,
where we analyzed what words occur in the same
sentence as other specific words. Using this information, we can estimate what words strongly bias
towards a category word, as well as calculate the
general sentiment of surrounding words.
We focused our analysis on gender, religion, and
race. Our goal is to provide users of this dataset
with preliminary guidance on how the different
components are biased so that they can make decisions on which components to train on.
All tables and figures in this section can be found
in the Appendix.
6.4.1 Gender
We computed gender associations by computing cooccurrences for binary pronouns. For each word,
we computed the difference in the rate it co-occurs
with "he" and "she"14and weighed it by the square
root of its frequency. We report the top 15 most
biased adjectives or adverbs (Loper and Bird, 2002)
for each in Table 10. We see that words like “military”, “criminal”, and “offensive” strongly bias
towards men, while “little”, “married”, “sexual”,
and “happy” bias towards women.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
43
|
towards men, while “little”, “married”, “sexual”,
and “happy” bias towards women.
In addition, we computed the average sentiment (Baccianella et al., 2010) of words cooccurring with the gendered pronouns across each
dataset in Figure 13. Generally, we find no significant sentiment bias towards men or women.
This, of course, does not mean that the dataset
is free of gender bias (as our co-occurrence tests
show).
6.4.2 Religion
We computed a similar co-occurrence analysis for
religion, which can be found in Table 11. Like gender, we find that these co-occurrences reflect how
these terms are used in pockets of online discourse.
For example, “radical” co-occurs with “muslim” at
a high rate, while “rational” often co-occurs with
“atheist”. This analysis also demonstrates some
of the limitations of a purely co-occurrence based
analysis. For example, “religious” often co-occurs
with “atheist”, which likely reflects the type of conversations in which the word “atheist” is likely to
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
44
|
with “atheist”, which likely reflects the type of conversations in which the word “atheist” is likely to
occur as opposed to a descriptor of “atheist”.
14We chose to only study male and female pronouns as a
simplifying assumption. Studying “they” would require us to
isolate its usage as a singular noun.
12
Figure 7: Log perplexity of 16-topic LDA trained on Pile-CC, on other Pile components. Dotted line indicates log
perplexity of the topic model on OpenWebText2. Higher indicates a larger topical divergence from Pile-CC.
Figure 8: Percentage of words classified as profane in
the Pile. The percentage of the CC component and the
weighted mean of the Pile as a whole are shown as horizontal lines.
In addition, we computed the average sentiment
of co-occurrences across each of the constituent
datasets in Figure 14. Over the entire dataset,
we find that “Buddhist” has the highest sentiment,
followed by “Hindu”, “Christian”, “Atheist”, and
“Muslim”. Notably, “Jew” is the lowest, perhaps
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
45
|
“Muslim”. Notably, “Jew” is the lowest, perhaps
reflecting its historical use as a pejorative.
6.4.3 Race
Finally, we ran the same analysis for racial groups.
Here, as identifiers like “black” or “white” often do not indicate race, we instead compute co-occurences with phrases like “black man” or “white
woman”.
We show the top 15 most biased words for each
demographic in Table 12. Once again, we found
that the co-occurrences reflect the context in which
these terms are used. For example, the 4 most
biased words for “black” are “unarmed”, “civil”,
“criminal”, and “scary”.
Similar to above, we compute the average sentiment of co-occurring words. We report the average
sentiment numbers in Table 13. We find that “hispanic/latino” narrowly edges out “asian” for the
highest sentiment, followed by “white”. On the
other hand, “black” had the lowest sentiment, at
-0.15.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
46
|
highest sentiment, followed by “white”. On the
other hand, “black” had the lowest sentiment, at
-0.15.
We note that for all demographics, the average sentiment is negative. We hypothesize that this is due
to the specific context for which the phrases we use
to compute co-occurrences appear. For example, it
is often quite common for news articles to describe
suspects as an “asian man”.
6.5 Author Consent and Public Data
Another issue with the use of texts in natural language processing research is consent. Although
one is typically not legally obligated to receive the
permission of an author to train a NLP algorithm on
their work15, many consider doing so a moral obli15Laws vary by country. For a discussion of US law, see
Section 7.1
13
gation or a good measure to guard against misuse
(Obar, 2020; Prabhu and Birhane, 2020). On the
other hand, there is significant disagreement surrounding the ethics of repurposing data protected
byterms of service in research contexts (Vitak et al.,
2016; Fiesler et al., 2020), particularly given the
power asymmetries inherent in digital platforms,
which often close off independent researchers from
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
47
|
2016; Fiesler et al., 2020), particularly given the
power asymmetries inherent in digital platforms,
which often close off independent researchers from
investigating public data while simultaneously compelling users to consent to its private use (Halavais,
2019).
While much of the Pile’s data comes from sources
that have expressly consented to its wider dissemination and use in research, researchers often fail to
clearly document where their data came from and
under what terms its use was consented to. In light
of this, we felt it appropriate to release the Pile with
transparency around how the authors of its data
have indicated that that data can be used.
To provide needed nuance to our discussion of consent, we identified three tiers of availability for
public use. Public data is data which is freely and
readily available on the internet. This primarily
excludes data which is pay-walled (regardless of
how easy that paywall is to bypass) and data which
cannot be easily obtained but can be obtained, e.g.
through a torrent or on the dark web. Terms of
Service (ToS) compliant data is data which is obtained and used in a fashion that is known to be
consistent with the terms of service of the data host.
Data with authorial consent is data for which the
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
48
|
consistent with the terms of service of the data host.
Data with authorial consent is data for which the
original authors of the work consented to the use
of their data, or where a reasonable person could
not assume that their data would not be used for
purposes such as research. ToS compliant data and
authorial consented data differ in two main ways:
It is important to keep in mind that people typically
do not read Terms of Service, and additionally that
being ToS-compliant does not entail authorial consent. We adopted a strict model of consent, where
ambiguous or unknown consent is treated as nonconsensual.
Table 5 summarizes our understanding of the status
of each of the datasets within the Pile. Datasets
marked with a 3are compliant in the relevant respects, though a couple datasets are worth remarking on in particular. Book3 and OpenSubtitles are
being used in a fashion that is consistent with the
terms of service of the data host . However, this is
somewhat misleading in that the data host is notauthorized to post the data online by the parties
that own it. The Enron Emails dataset was not
collected with the permission of the authors, but
was collected by the U.S. government as part of
a criminal investigation. While the people whose
emails are in the Enron dataset are aware of this
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
49
|
was collected by the U.S. government as part of
a criminal investigation. While the people whose
emails are in the Enron dataset are aware of this
fact, they were not given the ability to consent to
its inclusion in any way.
There are five datasets included in the Pile that were
not collected and distributed in a ToS compliant
fashion and for which the authors had no ability
to consent to their data being used. Each of these
datasets are widely used, both in the NLP literature and the world at large. With the exception
of the YouTube Subtitles dataset, each of these
datasets were published by researchers and are
passed around freely on the internet. The YouTube
Subtitles dataset was created by us for this project,
using a very popular unofficial API that is both
widely used and easily obtainable on Pip, Conda,
and GitHub, among other places. Given the processing applied and the difficulty of identifying particular files in the Pile, we feel that our use of these
datasets does not constitute significantly increased
harm beyond that which has already been done by
the widespread publication of these datasets.
7 Implications and Broader
Impacts
The Pile represents yet another stepping stone
along the path of scaling models and datasets to
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
50
|
the widespread publication of these datasets.
7 Implications and Broader
Impacts
The Pile represents yet another stepping stone
along the path of scaling models and datasets to
ever larger sizes and capabilities. There are many
serious concerns about how the emergence of progressively stronger AI systems will influence the
wider world (Brundage et al., 2018; Amodei et al.,
2016; Bostrom and Yudkowsky, 2014; Bostrom,
2014; Critch and Krueger, 2020), and we believe
that they merit serious thought. In this section
we discuss the legal ramifications of the Pile, and
then consider the impact of the Pile to AI alignment from two angles: accelerating AI timelines
and the dangers posed by unaligned language models.
7.1 Legality of Content
While the machine learning community has begun to discuss the issue of the legality of training
models on copyright data, there is little acknowledgment of the fact that the processing and distribution of data owned by others may also be a
violation of copyright law. As a step in that direc14
Component Public ToS Author
Pile-CC 3 3
PMC 3 3 3
Books3 3
OWT2 3
ArXiv 3 3 3
Github 3 3
FreeLaw 3 3 3
Stack Exchange 3 3 3
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
51
|
PMC 3 3 3
Books3 3
OWT2 3
ArXiv 3 3 3
Github 3 3
FreeLaw 3 3 3
Stack Exchange 3 3 3
USPTO 3 3 3
PubMed 3 3 3
PG-19 3 3
OpenSubtitles 3
Wikipedia 3 3 3
DM Math 3 3 3
Ubuntu IRC 3 3 3
BookCorpus2 3
EuroParl 3 3 3
HackerNews 3 3
YTSubtitles 3
PhilPapers 3 3 3
NIH 3 3 3
Enron Emails 3 3
Table 5: Types of consent for each dataset
tion, we discuss the reasons we believe that our
use of copyright data is in compliance with US
copyright law.16
Under pre (1984) (and affirmed in subsequent
rulings such as aff (2013); Google (2015)), noncommercial, not-for-profit use of copyright media
is preemptively fair use. Additionally, our use is
transformative , in the sense that the original form
of the data is ineffective for our purposes and our
form of the data is ineffective for the purposes of
the original documents. Although we use the full
text of copyright works, this is not necessarily disqualifying when the full work is necessary (ful,
2003). In our case, the long-term dependencies in
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
52
|
text of copyright works, this is not necessarily disqualifying when the full work is necessary (ful,
2003). In our case, the long-term dependencies in
natural language require that the full text be used in
order to produce the best results (Dai et al., 2019;
Rae et al., 2019; Henighan et al., 2020; Liu et al.,
2018).
Copyright law varies by country, and there may be
16This discussion does not, and is not intended to, constitute
legal advice; rather, it is a general discussion of law. Only your
attorney can provide assurances that the information contained
herein is applicable or appropriate to a particular situation. If
in doubt, it is always advisable to speak to an intellectual
property attorney.additional restrictions on some of these works in
particular jurisdictions. To enable easier compliance with local laws, the Pile reproduction code is
available and can be used to exclude certain components of the Pile which are inappropriate for
the user. Unfortunately, we do not have the metadata necessary to determine exactly which texts are
copyrighted, and so this can only be undertaken at
the component level. Thus, this should be be taken
to be a heuristic rather than a precise determination.
7.2 Acceleration of AI Timelines
There is serious concern that AI systems may soon
be meaningfully more capable than humans in all
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
53
|
7.2 Acceleration of AI Timelines
There is serious concern that AI systems may soon
be meaningfully more capable than humans in all
relevant economic tasks (Grace et al., 2018; Yudkowsky, 2013). Relatedly, there are serious unresolved questions surrounding how to properly align
such powerful AI systems with human interests
(Bostrom and Yudkowsky, 2014; Russell, 2019;
Bostrom, 2014; Amodei et al., 2016) and generally
avoid morally catastrophic outcomes (Sotala and
Gloor, 2017; Shulman and Bostrom, 2020). As
such, it has been argued that accelerating the development of such powerful AI systems may be
undesirable before these concerns have been more
adequately addressed (Bostrom, 2014).
There are several pragmatic responses to this
view:
1.Due to human competition, curiosity, and cultural diversity, halting technological development is incredibly difficult, if not impossible.
(Russell, 2019) (Critch and Krueger, 2020)
2.AI development is experimental in nature:
The alignment problem can only be solved
through development, testing and (hopefully
non-existential) failure.
3.High powered language models, along with
their more general successors, must be capable of viewing morally problematic content
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
54
|
through development, testing and (hopefully
non-existential) failure.
3.High powered language models, along with
their more general successors, must be capable of viewing morally problematic content
without adopting it in their output. We elaborate on this in the following section.
With this in mind, we accept the reality that the Pile
could potentially accelerate AI timelines. However,
we hope our efforts to establish best practices, such
as thoroughly documenting the contents of our data,
will help encourage diligence for downstream researchers on alignment problems.
15
7.3 Negative LM Output
There has been much discussion about the possible negative effects of powerful language models
in the world (Brown et al., 2020; Brundage et al.,
2018). Some of these possible problems, such as
the ability to mass produce low quality content
for the purpose of Search Engine Optimization,
are inherent problems to the way online content
is distributed, and cannot be stopped by those developing language models alone. Directly solving
these problems would require sweeping changes
to the architecture of the Internet, such as vastly
expanded Public Key Infrastructure and distributed
authentication of identity (Ferguson and Schneier,
2003).
Another concern is that training such models on
huge datasets will almost inevitably require them to
have undesirable content in their training sets, such
as that promoting hateful stereotypes (Christian,
2020). Having models output undesirable content
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
55
|
huge datasets will almost inevitably require them to
have undesirable content in their training sets, such
as that promoting hateful stereotypes (Christian,
2020). Having models output undesirable content
is, by definition, undesirable, but we believe that
attacking this problem from the training set side
is unproductive and ultimately leads us away from
optimal solutions. If a person reads a racist piece
of content, they do not then immediately adopt its
racist views—they may be capable of doing so, but
can decide not to. This capacity to understand undesirable content and then decide to ignore it is an
essential future research direction. Not only would
this allow models to use “dirtier" data with less
concern, but also to use their gained knowledge
to better understand what not to do. We recognize
that, despite recent progress in human-guided learning (Stiennon et al., 2020), the technology is not
yet at this stage, and have thus made a number of
editorial decisions as described in this paper. However, this approach seems essential to the future
of these models and AI more broadly, and more
research is needed.
8 Related Work
Self-supervised training of natural language processing models on large, unlabeled text corpora,
has seen widespread adoption in the field. Word
representation models such as GloVe (Pennington
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
56
|
has seen widespread adoption in the field. Word
representation models such as GloVe (Pennington
et al., 2014) and word2vec (Mikolov et al., 2013)
were trained on datasets such as Wikipedia, Gigaword (Graff et al., 2003), or a non-public Google
News corpus. More recently, language models
(Radford et al., 2018, 2019; Brown et al., 2020;Rosset, 2019; Shoeybi et al., 2019) and masked language models (Devlin et al., 2019; Liu et al., 2019;
Raffel et al., 2019) have been trained on datasets
such as Wikipedia, BookCorpus (Zhu et al., 2015),
RealNews (Zellers et al., 2019), CC-Stories (Trinh
and Le, 2018), and other Internet scrape-derived
datasets discussed below. Other datasets such as
WikiText (Stephen et al., 2016) have also been used
in similar self-supervised training.
As data requirements for language modeling have
grown, the field has turned towards Internet scrapes
for large-scale datasets (Gokaslan and Cohen,
2019), with Common Crawl being particularly
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
57
|
grown, the field has turned towards Internet scrapes
for large-scale datasets (Gokaslan and Cohen,
2019), with Common Crawl being particularly
prevalent. Works such as Brown et al. (2020); Wenzek et al. (2019); Suárez et al. (2019b); Raffel et al.
(2019) have relied on Common Crawl to build training datasets for large-scale models. However, these
works often highlight the difficulty of cleaning and
filtering the Common Crawl data, and often highlight the resulting data quality as a determining
factor of model capability.
It has also been increasingly common practice
to combine multiple datasets when training language models. For instance, GPT (Radford et al.,
2018) was trained on Wikipedia and BookCorpus,
whereas GPT-3 (Brown et al., 2020) was trained
on Wikipedia, two fiction datasets, and two webscraped datasets. The Pile continues the trend of
combining large-scale web-scrapes with smaller,
higher-quality datasets that capture knowledge we
believe would be most beneficial to training language models.
The two most comparable publicly available
datasets to the Pile are CC-100 (Wenzek et al.,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
58
|
The two most comparable publicly available
datasets to the Pile are CC-100 (Wenzek et al.,
2019) and C4/mC4 (Raffel et al., 2019). C4 is
comparably-sized to the Pile, while mC4 and CC100 are larger, multilingual datasets. However,
C4/mC4 require immense computational resources
to preprocess the data, with its maintainers even recommending the use of a distributed cloud service,17
setting a high bar of entry to using these datasets.
CC-100 is directly downloadable and pre-cleaned;
however, its English portion is much smaller than
the Pile. Importantly, these three datasets are all derived entirely from Common Crawl—as discussed
above, the current best practice in training largescale language models involve using both large web
scrapes and more targeted, higher-quality datasets,
17https://www.tensorflow.org/datasets/
catalog/c4
16
which the Pile directly addresses.
9 Acknowledgments
The authors would like to thank TensorFlow Research Cloud for providing the computational resources for the evaluation and OpenAI for providing access and credits for the OpenAI API for GPT3 evaluation.
We would also like to thank Farrukh Raman, JR
Smith, and Michelle Schmitz for reviewing the
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
59
|
We would also like to thank Farrukh Raman, JR
Smith, and Michelle Schmitz for reviewing the
manuscript.
References
1984. Sony corp. of america v. universal city studios,
inc.
2003. Kelly v. arriba soft corp.
2013. Righthaven llc v. hoehn.
Dario Amodei, Chris Olah, Jacob Steinhardt, Paul
Christiano, John Schulman, and Dan Mané. 2016.
Concrete problems in AI safety. arXiv preprint
arXiv:1606.06565 .
Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani. 2010. Sentiwordnet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining. In
LREC . European Language Resources Association.
Emily M Bender and Batya Friedman. 2018. Data
statements for natural language processing: Toward
mitigating system bias and enabling better science.
Transactions of the Association for Computational Linguistics , 6:587–604.
Stella Biderman. 2021. Data statement for the Pile.
arXiv preprint arXiv .
Stella Biderman, Kieran Bicheno, and Leo Gao. 2021.
Datasheet for the Pile. arXiv preprint arXiv .
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
60
|
Stella Biderman, Kieran Bicheno, and Leo Gao. 2021.
Datasheet for the Pile. arXiv preprint arXiv .
Stella Biderman and Walter J. Scheirer. 2020. Pitfalls
in machine learning research: Reexamining the development cycle. NeurIPS “I Can’t Believe It’s Not Better!” Workshop .
David M Blei, Andrew Y Ng, and Michael I Jordan.
2003. Latent dirichlet allocation. Journal of machine
Learning research , 3(Jan):993–1022.
Su Lin Blodgett, Solon Barocas, Hal Daumé III, and
Hanna Wallach. 2020. Language (technology) is
power: A critical survey of “bias” in NLP. arXiv
preprint arXiv:2005.14050 .
Nick Bostrom. 2014. Superintelligence: Paths, Dangers, Strategies . Oxford University Press, Inc.
Nick Bostrom and Eliezer Yudkowsky. 2014. The
ethics of artificial intelligence. The Cambridge handbook of artificial intelligence , 1:316–334.Tom B Brown, Benjamin Mann, Nick Ryder, Melanie
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
61
|
Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell,
et al. 2020. Language models are few-shot learners.
arXiv preprint arXiv:2005.14165 .
Miles Brundage, Shahar Avin, Jack Clark, Helen
Toner, Peter Eckersley, Ben Garfinkel, Allan Dafoe,
Paul Scharre, Thomas Zeitzoff, Bobby Filar, et al.
2018. The malicious use of artificial intelligence: Forecasting, prevention, and mitigation. arXiv preprint
arXiv:1802.07228 .
Nicholas Carlini, Florian Tramer, Eric Wallace,
Matthew Jagielski, Ariel Herbert-V oss, Katherine Lee,
Adam Roberts, Tom Brown, Dawn Song, Ulfar Erlingsson, Alina Oprea, and Colin Raffel. 2020. Extracting
training data from large language models.
Isaac Caswell, Theresa Breiner, Daan van Esch, and
Ankur Bapna. 2020. Language id in the wild: Unexpected challenges on the path to a thousand-language
web text corpus. arXiv preprint arXiv:2010.14571 .
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
62
|
web text corpus. arXiv preprint arXiv:2010.14571 .
Brian Christian. 2020. The Alignment Problem: Machine Learning and Human Values . WW Norton &
Company.
Alina Maria Ciobanu, Liviu P Dinu, and Andrea Sgarro.
2017. Towards a map of the syntactic similarity of languages. In International Conference on Computational
Linguistics and Intelligent Text Processing , pages 576–
590. Springer.
Alexis Conneau, Kartikay Khandelwal, Naman Goyal,
Vishrav Chaudhary, Guillaume Wenzek, Francisco
Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer,
and Veselin Stoyanov. 2020. Unsupervised crosslingual representation learning at scale. In Proceedings
of the 58th Annual Meeting of the Association for Computational Linguistics , pages 8440–8451, Online. Association for Computational Linguistics.
Marta R Costa-jussà, Roger Creus, Oriol Domingo,
Albert Domínguez, Miquel Escobar, Cayetana López,
Marina Garcia, and Margarita Geleta. 2020. Mtadapted datasheets for datasets: Template and repository. arXiv preprint arXiv:2005.13156 .
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
63
|
Andrew Critch and David Krueger. 2020. AI Research Considerations for Human Existential Safety
(ARCHES). Preprint at acritch.com/arches .
Zihang Dai, Zhilin Yang, Yiming Yang, Jaime
Carbonell, Quoc V Le, and Ruslan Salakhutdinov.
2019. Transformer-xl: Attentive language models beyond a fixed-length context. arXiv preprint
arXiv:1901.02860 .
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and
Kristina Toutanova. 2019. BERT: Pre-training of deep
bidirectional transformers for language understanding.
InProceedings of the 2019 Conference of the North
American Chapter of the Association for Computational Linguistics: Human Language Technologies,
Volume 1 (Long and Short Papers) . Association for
Computational Linguistics.
17
István Endrédy and Attila Novák. 2013. More effective boilerplate removal – the GoldMiner algorithm. In
Polibits .
Niels Ferguson and Bruce Schneier. 2003. Practical
Cryptography . John Wiley & Sons.
Casey Fiesler, Nathan Beard, and Brian C Keegan.
2020. No robots, spiders, or scrapers: Legal and ethical
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
64
|
Cryptography . John Wiley & Sons.
Casey Fiesler, Nathan Beard, and Brian C Keegan.
2020. No robots, spiders, or scrapers: Legal and ethical
regulation of data collection methods in social media
terms of service. In Proceedings of the International
AAAI Conference on Web and Social Media , volume 14,
pages 187–196.
Timnit Gebru, Jamie Morgenstern, Briana Vecchione,
Jennifer Wortman Vaughan, Hanna Wallach, Hal
Daumé III, and Kate Crawford. 2018. Datasheets for
datasets. arXiv preprint arXiv:1803.09010 .
Aaron Gokaslan and Vanya Cohen. 2019. Openwebtext corpus. http://Skylion007.github.io/
OpenWebTextCorpus .
Authors Guild v. Google. 2015. . Docket No. 13-4829cv, 804:202.
Katja Grace, John Salvatier, Allan Dafoe, Baobao
Zhang, and Owain Evans. 2018. When will AI exceed
human performance? evidence from AI experts. Journal of Artificial Intelligence Research , 62:729–754.
David Graff, Junbo Kong, Ke Chen, and Kazuaki
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
65
|
David Graff, Junbo Kong, Ke Chen, and Kazuaki
Maeda. 2003. English gigaword. Linguistic Data Consortium, Philadelphia , 4(1):34.
Declan Groves and Andy Way. 2006. Hybridity in mt:
Experiments on the Europarl corpus. In Proceeedings
of the 11th Annual conference of the European Association for Machine Translation (EAMT 2006) .
Alexander Halavais. 2019. Overcoming terms of service: a proposal for ethical distributed research. Information, Communication & Society , 22(11):1567–1581.
Chris Hardin. 2018. How to shuffle a big dataset.
Tom Henighan, Jared Kaplan, Mor Katz, Mark Chen,
Christopher Hesse, Jacob Jackson, Heewoo Jun, Tom B
Brown, Prafulla Dhariwal, Scott Gray, et al. 2020. Scaling laws for autoregressive generative modeling. arXiv
preprint arXiv:2010.14701 .
Matthew Hoffman, Francis Bach, and David Blei.
2010. Online learning for latent dirichlet allocation.
advances in neural information processing systems ,
23:856–864.
Dirk Hovy and Shannon L Spruit. 2016. The social impact of natural language processing. In Proceedings of
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
66
|
advances in neural information processing systems ,
23:856–864.
Dirk Hovy and Shannon L Spruit. 2016. The social impact of natural language processing. In Proceedings of
the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers) , pages
591–598.
Ben Hutchinson, Vinodkumar Prabhakaran, Emily
Denton, Kellie Webster, Yu Zhong, and Stephen Denuyl. 2020. Social biases in NLP models as barriers for persons with disabilities. arXiv preprint
arXiv:2005.00813 .Eun Seo Jo and Timnit Gebru. 2020. Lessons from
archives: Strategies for collecting sociocultural data in
machine learning. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency ,
pages 306–316.
Jared Kaplan, Sam McCandlish, Tom Henighan,
Tom B Brown, Benjamin Chess, Rewon Child, Scott
Gray, Alec Radford, Jeffrey Wu, and Dario Amodei.
2020. Scaling laws for neural language models. arXiv
preprint arXiv:2001.08361 .
Bryan Klimt and Yiming Yang. 2004. The Enron corpus: A new dataset for email classification research.
InEuropean Conference on Machine Learning , pages
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
67
|
Bryan Klimt and Yiming Yang. 2004. The Enron corpus: A new dataset for email classification research.
InEuropean Conference on Machine Learning , pages
217–226. Springer.
Sosuke Kobayashi. 2018. Homemade bookcorpus. https://github.com/BIGBALLON/
cifar-10-cnn .
Philipp Koehn. 2005. Europarl: A parallel corpus for
statistical machine translation. In MT summit , volume 5, pages 79–86. Citeseer.
Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu,
Dehao Chen, Orhan Firat, Yanping Huang, Maxim
Krikun, Noam Shazeer, and Zhifeng Chen. 2020.
Gshard: Scaling giant models with conditional computation and automatic sharding. arXiv preprint
arXiv:2006.16668 .
Peter J Liu, Mohammad Saleh, Etienne Pot, Ben
Goodrich, Ryan Sepassi, Lukasz Kaiser, and Noam
Shazeer. 2018. Generating wikipedia by summarizing
long sequences. arXiv preprint arXiv:1801.10198 .
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
68
|
Shazeer. 2018. Generating wikipedia by summarizing
long sequences. arXiv preprint arXiv:1801.10198 .
Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke
Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa:
A robustly optimized BERT pretraining approach.
arXiv preprint arXiv:1907.11692 .
Edward Loper and Steven Bird. 2002. NLTK: The Natural Language Toolkit. In Proceedings of the ACL
Workshop on Effective Tools and Methodologies for
Teaching Natural Language Processing and Computational Linguistics , pages 62–69. Somerset, NJ: Association for Computational Linguistics. http://arXiv.
org/abs/cs/0205028 .
John MacFarlane. 2006–2020. Pandoc: a universal
document converter.
Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. 2013. Distributed representations
of words and phrases and their compositionality. In Advances in Neural Information Processing Systems , volume 26, pages 3111–3119. Curran Associates, Inc.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
69
|
of words and phrases and their compositionality. In Advances in Neural Information Processing Systems , volume 26, pages 3111–3119. Curran Associates, Inc.
Jonathan A Obar. 2020. Sunlight alone is not a disinfectant: Consent and the futility of opening big data
black boxes (without assistance). Big Data & Society ,
7(1):2053951720935615.
Jeffrey Pennington, Richard Socher, and Christopher D.
Manning. 2014. Glove: Global vectors for word repre18
sentation. In Empirical Methods in Natural Language
Processing (EMNLP) , pages 1532–1543.
Vinay Uday Prabhu and Abeba Birhane. 2020. Large
image datasets: A pyrrhic win for computer vision?
arXiv preprint arXiv:2006.16923 .
Shawn Presser. 2020. Books3. https:
//twitter.com/theshawwn/status/
1320282149329784833 .
Alec Radford, Karthik Narasimhan, Time Salimans,
and Ilya Sutskever. 2018. Improving language understanding with unsupervised learning. Technical report,
OpenAI .
Alec Radford, Jeffrey Wu, Rewon Child, David Luan,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
70
|
OpenAI .
Alec Radford, Jeffrey Wu, Rewon Child, David Luan,
Dario Amodei, and Ilya Sutskever. 2019. Language
models are unsupervised multitask learners. OpenAI
Blog , 1(8):9.
Jack W Rae, Anna Potapenko, Siddhant M Jayakumar,
Chloe Hillier, and Timothy P Lillicrap. 2019. Compressive transformers for long-range sequence modelling.
arXiv preprint .
Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou,
Wei Li, and Peter J Liu. 2019. Exploring the limits of
transfer learning with a unified text-to-text transformer.
arXiv preprint arXiv:1910.10683 .
Inioluwa Deborah Raji and Jingying Yang. 2019.
ABOUT ML: Annotation and benchmarking on understanding and transparency of machine learning lifecycles. arXiv preprint arXiv:1912.06166 .
C. Radhakrishna Rao. 1961. Generation of random permutations of given number of elements using random
sampling numbers. Sankhy ¯a: The Indian Journal of
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
71
|
C. Radhakrishna Rao. 1961. Generation of random permutations of given number of elements using random
sampling numbers. Sankhy ¯a: The Indian Journal of
Statistics, Series A (1961-2002) , 23(3):305–307.
Radim Rehurek, Petr Sojka, et al. 2011. Gensim—statistical semantics in python. NLP Centre, Faculty of Informatics, Masaryk University .
C Rosset. 2019. Turing-NLG: A 17-billion-parameter
language model by Microsoft. Microsoft Blog .
S. Russell. 2019. Human Compatible: Artificial Intelligence and the Problem of Control . Penguin Publishing
Group.
David Saxton, Edward Grefenstette, Felix Hill, and
Pushmeet Kohli. 2019. Analysing mathematical reasoning abilities of neural models. arXiv preprint
arXiv:1904.01557 .
Ismaïla Seck, Khouloud Dahmane, Pierre Duthon,
and Gaëlle Loosli. 2018. Baselines and a datasheet
for the Cerema AWP dataset. arXiv preprint
arXiv:1806.04016 .
Mohammad Shoeybi, Mostofa Patwary, Raul Puri,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
72
|
arXiv:1806.04016 .
Mohammad Shoeybi, Mostofa Patwary, Raul Puri,
Patrick LeGresley, Jared Casper, and Bryan Catanzaro.
2019. Megatron-LM: Training multi-billion parameter
language models using gpu model parallelism. arXiv
preprint arXiv:1909.08053 .Carl Shulman and Nick Bostrom. 2020. Sharing the
world with digital minds. preprint .
Kaj Sotala and Lukas Gloor. 2017. Superintelligence
as a cause or cure for risks of astronomical suffering.
Informatica , 41(4).
Robyn Speer. 2019. ftfy. Zenodo. Version 5.5.
Merity Stephen, Xiong Caiming, Bradbury James, and
Richard Socher. 2016.
Nisan Stiennon, Long Ouyang, Jeffrey Wu, Daniel
Ziegler, Ryan Lowe, Chelsea V oss, Alec Radford,
Dario Amodei, and Paul F Christiano. 2020. Learning to summarize with human feedback. Advances in
Neural Information Processing Systems , 33.
Pedro Javier Ortiz Suárez, Benoît Sagot, and Laurent
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
73
|
Neural Information Processing Systems , 33.
Pedro Javier Ortiz Suárez, Benoît Sagot, and Laurent
Romary. 2019a. Asynchronous pipeline for processing huge corpora on medium to low resource infrastructures. In 7th Workshop on the Challenges in the Management of Large Corpora (CMLC-7) . Leibniz-Institut
für Deutsche Sprache.
Pedro Javier Ortiz Suárez, Benoît Sagot, and Laurent
Romary. 2019b. Asynchronous pipeline for processing huge corpora on medium to low resource infrastructures. In 7th Workshop on the Challenges in the Management of Large Corpora (CMLC-7) . Leibniz-Institut
für Deutsche Sprache.
Anja Thieme, Danielle Belgrave, and Gavin Doherty.
2020. Machine learning in mental health: A systematic review of the HCI literature to support the development of effective and implementable ML systems.
ACM Transactions on Computer-Human Interaction
(TOCHI) , 27(5):1–53.
J. Tiedemann. 2016. Finding alternative translations in
a large corpus of movie subtitles. Proceedings of the
10th International Conference on Language Resources
and Evaluation (LREC 2016) .
Trieu H. Trinh and Quoc V . Le. 2018. A simple method
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
74
|
10th International Conference on Language Resources
and Evaluation (LREC 2016) .
Trieu H. Trinh and Quoc V . Le. 2018. A simple method
for commonsense reasoning. CoRR , abs/1806.02847.
Hans Van Halteren. 2008. Source language markers in
Europarl translations. In Proceedings of the 22nd International Conference on Computational Linguistics
(Coling 2008) , pages 937–944.
Jessica Vitak, Katie Shilton, and Zahra Ashktorab.
2016. Beyond the Belmont principles: Ethical challenges, practices, and beliefs in the online data research
community. In Proceedings of the 19th ACM Conference on Computer-Supported Cooperative Work & Social Computing , pages 941–953.
Guillaume Wenzek, Marie-Anne Lachaux, Alexis Conneau, Vishrav Chaudhary, Francisco Guzmán, Armand
Joulin, and Edouard Grave. 2019. CCNet: Extracting
high quality monolingual datasets from web crawl data.
arXiv preprint arXiv:1911.00359 .
Thomas Wolf, Lysandre Debut, Victor Sanh, Julien
Chaumond, Clement Delangue, Anthony Moi, Pierric
Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
75
|
Chaumond, Clement Delangue, Anthony Moi, Pierric
Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe
19
Davison, Sam Shleifer, Patrick von Platen, Clara Ma,
Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao,
Sylvain Gugger, Mariama Drame, Quentin Lhoest, and
Alexander M. Rush. 2020. Transformers: State-ofthe-art natural language processing. In Proceedings of
the 2020 Conference on Empirical Methods in Natural
Language Processing: System Demonstrations , pages
38–45, Online. Association for Computational Linguistics.
Ellery Wulczyn, Nithum Thain, and Lucas Dixon. 2016.
Wikipedia detox.
Eliezer Yudkowsky. 2013. Intelligence explosion microeconomics. Machine Intelligence Research Institute, accessed online October , 23:2015.
Rowan Zellers, Ari Holtzman, Hannah Rashkin,
Yonatan Bisk, Ali Farhadi, Franziska Roesner, and
Yejin Choi. 2019. Defending against neural fake
news. In H. Wallach, H. Larochelle, A. Beygelzimer,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
76
|
Yejin Choi. 2019. Defending against neural fake
news. In H. Wallach, H. Larochelle, A. Beygelzimer,
F. dÁlché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems 32 ,
pages 9054–9065. Curran Associates, Inc.
Victor Zhou. 2019. Building a better profanity detection library with scikit-learn.
Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books and movies: Towards storylike visual explanations by watching movies and reading books. In Proceedings of the IEEE international
conference on computer vision , pages 19–27.
20
Appendices
A Contributions
All authors contributed to the design of the research
project and the writing of the paper. Additionally,
authors contributed as follows:
Leo Gao led the project, implemented the main
Pile codebase, contributed to the model training
code, performed the evaluations and the language
analysis, interpreted the perplexity analysis results,
implemented the processing to create the final data,
and processed Pile-CC, PubMed Central, ArXiv,
and Ubuntu IRC.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
77
|
implemented the processing to create the final data,
and processed Pile-CC, PubMed Central, ArXiv,
and Ubuntu IRC.
Stella Biderman led the data analysis, the broader
impact analysis, and the data documentation, and
coordinated the project. She also wrote the analysis of structural statistics, authorial consent, and
copyright law.
Sid Black implemented the model training and
evaluation code and processed YouTube Subtitles,
Stack Exchange, and GitHub.
Laurence Golding implemented deduplication,
performed the n-gram analysis, and processed
OpenWebText2.
Travis Hoppe processed FreeLaw, Pubmed Abstracts, ExPorter, and PhilPapers.
Charles Foster performed the topic modeling analysis, contributed to the discussion of authorial consent, and processed USPTO Backgrounds.
Jason Phang implemented and performed the GPT2/3 perplexity analysis and advised the project.
Horace He performed the bias and sentiment analysis.
Anish Thite implemented and performed the profanity analysis and processed Hacker News.
Noa Nabeshima processed GitHub.
Shawn Presser processed BookCorpus2.
Connor Leahy wrote the alignment implication
analysis and the model training code.
B Excluded Datasets
In the course of building the Pile, we considered
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
78
|
Connor Leahy wrote the alignment implication
analysis and the model training code.
B Excluded Datasets
In the course of building the Pile, we considered
including and ultimately decided to not use several datasets. We excluded several datasets on the
grounds that they were too small to be worth spending time on or because the English component of
the data did not merit inclusion on its own. However we also decided to exclude several data sets
for other reasons, which we document here fortransparency:
1.US Congressional Record. The official
record of the United States Congress ( 1800
– today) records important points of debate at
the highest levels of American government. It
reflects the opinions and biases of the political class over the past 200 years, including
segregationism and xenophobia. In particular,
we found a large quantity of extremely racist
content that we did not feel appropriate for a
dataset intended for general-purpose language
modeling.
2.Fanfiction. Hundreds of GiB of fanfiction
has been written and put online, primarily
on the websites www.fanfiction.net
andwww.https://archiveofourown.
org/ . This represents a significant untapped
resource for language modeling as it is almost exclusively short-form fiction, a writing
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
79
|
org/ . This represents a significant untapped
resource for language modeling as it is almost exclusively short-form fiction, a writing
style that is not represented in most language
modeling datasets. We ultimately decided to
exclude fanfiction on logistical grounds: we
found other sources of data that were easier to
obtain.
3.Literotica. Literotica is a website where users
can upload short-form erotic fiction. We had
originally planned on including it in the Pile
and even went as far as scraping and processing it. However we decided to not include it
for several reasons. Firstly, once we decided
to exclude fanfiction, Literotica represented
our sole source of short-form fiction, which
would likely lead to undesirable biases in the
trained model. Secondly, Literotica would
require significantly more investigation, assessment, and care than we spent on the other
datasets. Thirdly, Literotica contains a significant amount of stereotyping, including racial
fetishes. While Literotica is likely usable for
some tasks, we are not comfortable including
it in the Pile.
C Dataset Details
This section contains additional information about
each dataset listed in Section 2, including how it
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
80
|
some tasks, we are not comfortable including
it in the Pile.
C Dataset Details
This section contains additional information about
each dataset listed in Section 2, including how it
was obtained, how it was processed, and any other
details relevant for replication. The intent of this
section is to provide as much detail as possible,
so that Pile can be replicated in the future if necessary, and so that any future processing of these
21
and similar datasets can use or improve on our
methods. As such, all code created for processing
has been made publicly available under permissive
open source licenses and is referenced in footnotes
where applicable.
C.1 Pile-CC
We extract Common Crawl using jusText (Endrédy
and Novák, 2013). Our filtering implementation
uses a classifier trained against the OpenWebText2
dataset. We process only a small fraction of the
available Common Crawl data; we break the list
of urls to individual WARC files from 2013 to
2020 into 3679 chunks and process 22 random
chunks.
C.1.1 WARC vs WET
CommonCrawl data is available in two main formats: Web ARChive (WARC) files, which contain
a full record of the crawl as well as the raw HTML
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
81
|
CommonCrawl data is available in two main formats: Web ARChive (WARC) files, which contain
a full record of the crawl as well as the raw HTML
of the webpage, and WET files, which contain preextracted versions of the contents of the WARC
files. The WET files have poor quality, often containing large amounts of boilerplate text like menus
and page footers, but due to the lower bandwidth
and computation requirements necessary to use
WET files, prior work based on CC have mainly
focused on using WET files while applying cleaning such as document level filtering (Brown et al.,
2020; Wenzek et al., 2019), or n-sentence level
deduplication with very aggressive heuristics (Raffel et al., 2019).
We do not believe that document level filtering
is sufficient for WET files because many of the
issues with WET files stem from intra-document
boilerplate. We also find many of the heuristics
used in Raffel et al. (2019), such as the removal
of all lines without terminal punctuation, the word
"javascript", and 3-sentence deduplication to be too
aggressive.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
82
|
of all lines without terminal punctuation, the word
"javascript", and 3-sentence deduplication to be too
aggressive.
C.1.2 Extraction
In addition to jusText, we also considered Trafilatura, Newspaper, Goose3, and DragNet. While
we were originally intending on creating an extraction benchmark, this proved infeasible given our
available resources, and we chose jusText based on
visual inspection of the output. In inspection, we
noticed that jusText has the characteristic that it discards more data than many other extractors, which
is not a major drawback given the large volume
of CC data available. This was as expected, givenjusText’s intended application for text corpora creation. In contrast, trafilatura is, for instance, better
at preserving the structure of the website faithfully,
often correctly extracting elements such as tables,
but it kept too much unnecessary boilerplate. Had
we used trafilatura, we would have required an additional intra-page filtering step to remove boilerplate
from the page.
C.1.3 Languages
While jusText does technically support several
other languages, the quality on those languages
is worse than on English as many constants in the
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
83
|
from the page.
C.1.3 Languages
While jusText does technically support several
other languages, the quality on those languages
is worse than on English as many constants in the
algorithm are specifically tuned for English. Additionally, jusText is completely unable to handle
languages such as Chinese and Japanese, which do
not use spaces to delimit words.
Due to the difficulty of maintaining an acceptable
level of extraction quality across all languages, we
decided to restrict the scope of the CC dataset to
only English and leave a high-quality, fully multilingual, WARC-based CC-based dataset to future
work. To filter for only English, we use the pycld2 library and only attempt to extract text from
documents where English is the most common language.
We use pycld2 instead of fasttext because it is capable of classifying the language from the HTML
directly, and since jusText requires knowledge of
the language of the webpage before extraction. Additionally, pycld2 was significantly faster than jusText, and by only processing with jusText documents classified as English by pycld2, we reduced the required computation by approximately
half.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
84
|
half.
Extracting text from websites for language modeling, especially for multilingual corpora, is highly
nontrivial, and we leave the refinement of such
extraction to future work.
C.1.4 Filtering
To filter CC for quality, we follow Brown et al.
(2020) in training a classifier to classify between a
known high quality dataset and CC. We use fasttext
with an n-gram size of 2. We ran experiments using both the entire Pile and just OpenWebText2 as
the positive examples, with score distributions on
unseen CC data as shown in Figure 9. We decided
to use only OpenWebText2 for positive examples
for our final CC data because of the low sensitivity
22
Filtering Ratio
1 0.5894
2 0.3649
3 0.2390
4 0.1671
5 0.1239
6 0.0974
7 0.0802
8 0.0685
9 0.0602
Table 6: Filtering Ratios (kept:total) of various settings
of using the full Pile. We use the same Paretodistribution thresholding as Brown et al. (2020),
with = 3. Our choice of targets the filtering
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
85
|
with = 3. Our choice of targets the filtering
ratio necessary to filter our subset of CC to the size
we needed. The impact of on the filtering ratio is
shown in Table 6.
C.2 Pubmed Central
We use pandoc 1.19.2.4 (MacFarlane, 2006–
2020) to convert the JATS format data provided by
PMC to markdown. Afterwards, we remove any
line beginning with :::, which is used by pandoc
to indicate html classes in markdown.
C.3 Books3
No additional details.
C.4 OpenWebText2
To produce the dataset, URLs and their associated
metadata were first extracted from all Reddit submissions up to April 2020. URLs were deduplicated, with each unique URL featuring a list of
associated submissions metadata, and an aggregate score. URLs with an aggregate score of less
then 3 were removed. The links were then scraped
and processed with Newspaper scraper. Deduplication was performed at the document level using
in memory MinHashLSH through the DataSketch
library.
Both filtered and raw versions were produced, with
the raw version only deduplicated by URL. The filtered version contains 65.86 GB of uncompressed
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
86
|
Both filtered and raw versions were produced, with
the raw version only deduplicated by URL. The filtered version contains 65.86 GB of uncompressed
text across 17,103,059 documents. The raw version
is much larger, at 193.89GB of uncompressed text
across 69,547,149 documents.C.4.1 Extractor Choice
We chose to use Newspaper instead of jusText for
OpenWebText2 for consistency with OpenWebTextCorpus. Additionally, by using multiple different html extractors for different components of the
Pile, we reduce the potential impact of systematic
biases from any one extractor negatively impacting
the dataset.
C.5 ArXiv
We downloaded the T EX sources of all
papers on arXiv up to the July 2020
dump (the last file included in our data is
arXiv_src_2007_068.tar ) via arXiv’s S3
Bulk Source File Access18, and used pandoc
1.19.2.4 to convert these source files to
Markdown, discarding any papers which had errors
during the conversion process. This yielded a total
of 1,264,405 papers.
We remove any line beginning with :::, which is
used by pandoc to indicate html classes in markdown.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
87
|
of 1,264,405 papers.
We remove any line beginning with :::, which is
used by pandoc to indicate html classes in markdown.
C.6 GitHub
We separate the data gathering process into two
steps:
1.Gathering a list of the desired repositories and
their metadata
2.Extracting all text data useful for language
modeling from each repository
For the first step, mirroring the approach of the
WebText dataset, we use GitHub ‘stars’ as a proxy
for quality, and choose to gather only repositories
with more than 100stars. For practical reasons, we
also limit the list of repositories gathered to repositories with less than 1GB of files. Since Github’s
API limits the number of search results to 1000 , in
order to comprehensively gather all repositories we
need to create many small queries that each return
fewer than 1000 results in such a way that every
repository of interest will be returned by at least
one of our queries. To achieve this, we bound our
initial search by size to return only repositories between a lower bound of 0and5bytes. At the time
of writing, this returns 965results. For the next
step, we set our lower bound one above our previous upper bound, and decide on a new upper bound
that should also return fewer than 1000 results by
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
88
|
step, we set our lower bound one above our previous upper bound, and decide on a new upper bound
that should also return fewer than 1000 results by
18https://arxiv.org/help/bulk_data_s3
23
(a) OpenWebText2
(b) Full Pile
Figure 9: Score distribution of documents from Common Crawl given different classifier training data.
Figure 10: Left: number of new submissions/year
to arXiv grouped by domain over time. Right:
fractional submission rates for each of the domains.
Figure from https://arxiv.org/help/
stats/2019_by_area/
using the results from our last query to estimate our
new upper bound as (lowerbound+(1000 =(n=r)),
wherenis the number of previous results and ris
the range of bounds in the previous step.
This tends not to overshoot, because Github repositories follow a power distribution with respect to
size, but if it does, we simply use the amount of
repositories our new query returned in order to construct a new upper bound estimate.
Using the gathered list of repositories, we clone
each one, extract any text-based files, and discard
the rest. Because some repositories took an impractical amount of time to clone and/or extract, we set
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
89
|
each one, extract any text-based files, and discard
the rest. Because some repositories took an impractical amount of time to clone and/or extract, we set
a hard time limit of 300 seconds for both the git
cloning and text extraction steps. As such, some
larger repositories may only be partially extracted.
We also impose a file size limit of 100kB on extracted files, as we found that the majority of files
over that size were typically very repetitive autogenerated source files or data files, and that setting
this file size limit was an effective cleaning step to
limit the data to code.Because we wanted to limit the size of the overall
Pile, we randomly sampled 95.0 GiB of the 630.64
GiB of Github data we collected in total and leave
quality filtering to future work.
However, we believe code generation will be an increasingly important component of language models as they continue to scale up and increase in their
ability to generalize. As such, we hope to extend
this dataset in future work.
C.7 FreeLaw
We download the court opinions data in bulk from
CourtListener,19and extract the raw text using
BeautifulSoup .
C.8 Stack Exchange
To construct the dataset, we download and parse
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
90
|
CourtListener,19and extract the raw text using
BeautifulSoup .
C.8 Stack Exchange
To construct the dataset, we download and parse
every Stack Exchange database dump to plaintext
files. We opt to extract the top three answers
with at least three upvotes, discarding all other
responses. We only include the plain text question and response and do not incorporate any metadata. Motivated by large-scale language models’
few-shot ability (Brown et al., 2020), we provide
context by prepending all questions and answers
withQ:\n\n andA:\n\n respectively.
The resulting dataset contains a total of 15,622,475
documents across a total of 365 Stack Exchanges
and Meta-Stack Exchanges, the bulk of which is
from StackOverflow.
C.9 USPTO Backgrounds
The United States Patent and Trademark Office
(USPTO) has published bulk archives of the full
19https://www.courtlistener.com/api/
bulk-info/
24
text of all patents granted in the US from 1976 to
September 2020. From these archives, we extract
the Background sections, along with key grantspecific metadata, such as the inventor, assignee,
and classification information.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
91
|
the Background sections, along with key grantspecific metadata, such as the inventor, assignee,
and classification information.
The file format used for storing bulk text US patents
has changed over time. Prior to 2002, all of the
datasets are in a specialized format called APS
(Automated Patent System). Since 2002, the data
is XML encoded. Partially as a function of this
change, the location of the "Background" section
has also shifted. Our converter accounts for these
structural shifts and extracts the raw text from each
patent’s Background.
C.10 PubMed Abstracts
About one-third of the articles in the dataset were
missing or contained a malformed title or abstract
and were excluded. Additionally, PubMed Central (see Section 2.2) contains full-text resources to
many recent publications; any publications which
already appear in PMC are excluded from this set.
To process the data, we concatenated the title and
abstract and removed any copyright information.
The remaining dataset contains 15,518,009 titles
and abstracts.
C.11 Project Gutenberg
No additional details.
C.12 OpenSubtitles
To create the text dataset, we simply extract the
subtitle text from each XML file in the English
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
92
|
No additional details.
C.12 OpenSubtitles
To create the text dataset, we simply extract the
subtitle text from each XML file in the English
language dataset provided by Tiedemann (2016),
discarding any provided metadata.
C.13 Wikipedia (English)
We use the wikipedia/20200301.en dataset
from TensorFlow Datasets.20We prepend the title to the body of each article, separated by two
newlines.
C.14 DeepMind Mathematics
We include instances from the Easy, Medium,
and Hard components of DeepMind Mathematics, breaking each curriculum item (such as
algebra__polynomial_roots ) into 8 KiB
chunks.
20https://www.tensorflow.org/datasets/
catalog/wikipedia#wikipedia20200301enC.15 Ubuntu IRC
We processed all logs from July 5, 2004 through
September 1, 2020.
To process the data, all system messages, such
as joins, disconnects, nick changes, etc. were
discarded, but actions (i.e using /me) were kept.
Timestamps were removed, and all logs for the
same channel in a given week were concatenated
into a single document, with each the logs for each
day prepended with the date if that day’s log is
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
93
|
same channel in a given week were concatenated
into a single document, with each the logs for each
day prepended with the date if that day’s log is
non-empty.
C.16 BookCorpus2
The original BookCorpus consists of 11,038 books.
However, due to issues with availability of the
original BookCorpus, as well as the possibility of
collecting a larger version, we decided to collect
our own version of BookCorpus using a similar
methodology as Kobayashi (2018). Our version of
BookCorpus contains 17,868 books instead.
We create and use a modified version of the epubto-text converter in Kobayashi (2018) that:
•Correctly preserves the document structure
across chapters, matching the table of contents
very closely;
•Correctly renders tables of data, whereas by
default html2txt produces poor-quality results for tables,
•Correctly preserves code structure, so that
source code is visually coherent,
• Converts numbered lists from “1\.” to “1.”
•Runs the full text through
ftfy.fix_text() (Speer, 2019),
replacing Unicode apostrophes with ascii
apostrophes and expanding Unicode ellipses
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
94
|
ftfy.fix_text() (Speer, 2019),
replacing Unicode apostrophes with ascii
apostrophes and expanding Unicode ellipses
to “...” (three separate ascii characters).
C.17 EuroParl
We download the data in bulk from21. We remove all basic tag information and only retain
the name of each document as a title. For example, <SPEAKER ID=77 LANGUAGE="NL"
NAME="Pronk"> becomes Pronk , and then extract the body of each document, discarding those
that are shorter than 200 characters.
21http://www.statmt.org/europarl/
25
C.18 HackerNews
We first use the Hackernews BigQuery dataset to
obtain a list of all story ids in our date range. For
the Pile we use the first Hacker News post (1) to
post number 24531712. This corresponds to a date
range of approximately 10/09/2006 to 09/20/2020.
We use the BigQuery dataset to gather story ids
for efficiency purposes. However, the BigQuery
dataset was lacking some information for stories,
so we used the official Hacker News API for story
and comment text retrieval.
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
2101.00027
|
95
|
dataset was lacking some information for stories,
so we used the official Hacker News API for story
and comment text retrieval.
Hacker News displays and stores comments in a
tree-like manner, with children comments replying
to parent comments. However, most language models require input data to be in a sequential form.
Considering each path through the comment tree
as a sequence could be detrimental, since there will
be a large amount of near-duplicate comment sequences. In addition, only taking one path through
the comment tree for each story leaves out a large
portion of the comment data. Therefore, we parsed
comments in a hybrid form. For every top-level
comment (comments that have no parent comment),
we create a sequence of comments by traversing
down the comment tree from the top-level comment. We choose the next comment by taking the
child comment with the highest number of children
comments (a cheap attempt at taking a long path
through the comment tree, note that it does not take
the longest possible path).
We consider all stories that have at least one comment and are not flagged by the moderators for
potential conduct violations. Since comments are
stored in HTML, we use the html2text package
to extract the text from the post.
We order each document by listing the title, url,
|
2101.00027
|
The Pile: An 800GB Dataset of Diverse Text for Language Modeling
|
Recent work has demonstrated that increased training dataset diversity
improves general cross-domain knowledge and downstream generalization
capability for large-scale language models. With this in mind, we present
\textit{the Pile}: an 825 GiB English text corpus targeted at training
large-scale language models. The Pile is constructed from 22 diverse
high-quality subsets -- both existing and newly constructed -- many of which
derive from academic or professional sources. Our evaluation of the untuned
performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on
many of its components, such as academic writing. Conversely, models trained on
the Pile improve significantly over both Raw CC and CC-100 on all components of
the Pile, while improving performance on downstream evaluations. Through an
in-depth exploratory analysis, we document potentially concerning aspects of
the data for prospective users. We make publicly available the code used in its
construction.
|
http://arxiv.org/pdf/2101.00027
|
[
"Leo Gao",
"Stella Biderman",
"Sid Black",
"Laurence Golding",
"Travis Hoppe",
"Charles Foster",
"Jason Phang",
"Horace He",
"Anish Thite",
"Noa Nabeshima",
"Shawn Presser",
"Connor Leahy"
] |
[
"cs.CL"
] | null | null |
cs.CL
|
20201231
|
20201231
|
[
{
"id": "2101.00027"
},
{
"id": "1606.06565"
},
{
"id": "2005.14050"
},
{
"id": "2005.14165"
},
{
"id": "1802.07228"
},
{
"id": "2010.14571"
},
{
"id": "2005.13156"
},
{
"id": "1901.02860"
},
{
"id": "1803.09010"
},
{
"id": "2010.14701"
},
{
"id": "2005.00813"
},
{
"id": "2001.08361"
},
{
"id": "2006.16668"
},
{
"id": "1801.10198"
},
{
"id": "1907.11692"
},
{
"id": "2006.16923"
},
{
"id": "1910.10683"
},
{
"id": "1912.06166"
},
{
"id": "1904.01557"
},
{
"id": "1806.04016"
},
{
"id": "1909.08053"
},
{
"id": "1911.00359"
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.