Datasets:

Modalities:
Text
Formats:
json
ArXiv:
Libraries:
Datasets
Dask
License:
monology commited on
Commit
3f950f4
1 Parent(s): 743bf92

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -0
README.md CHANGED
@@ -1,3 +1,12 @@
1
  ---
2
  license: other
3
  ---
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: other
3
  ---
4
+
5
+ # Pile Uncopyrighted
6
+ In response to [authors demanding that LLMs stop using their works](https://tcrn.ch/3rtpIDn), here's a copy of [The Pile](https://huggingface.co/datasets/monology/pile) with all copyrighted content removed.
7
+ Please consider using this dataset to train your future LLMs, to respect authors and abide by copyright law.
8
+ Creating an uncopyrighted version of a larger dataset (ie RedPajama) is planned, with no ETA.
9
+
10
+ **Methodology**
11
+ Cleaning was performed by removing everything from the Books3, BookCorpus2, OpenSubtitles, YTSubtitles, and OWT2 subsets.
12
+ Based on section 7.1 of [the original paper](https://arxiv.org/abs/2101.00027), these datasets are the only ones which are not explicitly allowed to be used in AI training.