mirror of
https://github.com/facebookresearch/fairseq.git
synced 2024-09-11 09:15:43 +03:00
Summary: - Full unflattened optimizer state dict is in `checkpoints/shard_0.pt`, other checkpoint files do not have the `last_optimizer_state` key. - requires master version of fairscale (eventually fairscale>=0.3.3) Pull Request resolved: https://github.com/fairinternal/fairseq-py/pull/1744 Reviewed By: myleott Differential Revision: D27342305 Pulled By: sshleifer fbshipit-source-id: 7442b8c6ed01599d8ab0050213e84051f4e98acd |
||
---|---|---|
.. | ||
constraints | ||
__init__.py | ||
average_checkpoints.py | ||
build_sym_alignment.py | ||
compare_namespaces.py | ||
compound_split_bleu.sh | ||
convert_dictionary.lua | ||
convert_model.lua | ||
count_docs.py | ||
read_binarized.py | ||
rm_pt.py | ||
sacrebleu.sh | ||
shard_docs.py | ||
split_train_valid_docs.py | ||
spm_decode.py | ||
spm_encode.py | ||
spm_train.py | ||
test_fsdp.sh |