Datasets:

Modalities:
Text
Formats:
parquet
Libraries:
Datasets
Dask
License:

Are you considering releasing the semantically deduplicated data?

#2
by ricomnl - opened

Thanks for creating and releasing this amazing resource! As per my current understanding, the released data is not yet semantically deduplicated and I was wondering if you're considering releasing the processed, deduplicated version?

Tatta Bio org

Thanks! Yes that's correct the released data is quality filtered following the steps outlined in the OMG paper, before tokenization and semdedup. We are not currently planning to release a tokenized, semdedup version. The code for semdedup is available from FAIR here: https://github.com/facebookresearch/SemDeDup

andrecornman changed discussion status to closed

Sign up or log in to comment