facebookresearch / fairseq2

FAIR Sequence Modeling Toolkit 2
https://facebookresearch.github.io/fairseq2/
MIT License
678 stars 78 forks source link

adding a barrier so that tmp dir stays alive #606

Closed uralik closed 3 months ago

uralik commented 3 months ago

What does this PR do? Please describe:

ckpt saving was crashing since some ranks attempted to access model tmp dir when it was removed already

Fixes #{issue number}

Does your PR introduce any breaking changes? If yes, please list them: List of all backwards-incompatible changes.

Check list: