A fast, memory efficient Python module to remove duplicates from parallel text corpora.
It's useful for cleaning up datasets that contain duplicate entries for training language models.
pip install removedupfrom removedup import rdup
src, tgt, removed = rdup("source.txt", "target.txt")
print(src, tgt, removed)
# source.txt.dedup
# target.txt.dedup
# <num lines removed>Source and target must have the same number of lines. No validation checks are made.
Duplication checks are made both on the source and target content. E.g. duplication needs to appear on both source and target.
This is a breaking change from versions < 1.1.0.
git clone https://github.com/LibreTranslate/RemoveDup
cd RemoveDup
python setup.py buildYou can also use removedup as a standalone Windows, macOS or Linux application (but you currently need to build from source, we don't provide binaries).
mkdir build
cd build && cmake .. && make -j4
./rdup source.txt target.txt
We welcome pull requests!
AGPLv3