Don't reload best validation loss when using --reset-optimizer

Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/661 Differential Revision: D15068312 Pulled By: myleott fbshipit-source-id: 1216835fd4c7f83ea5e350bff83901c93ac57447

Don't reload best validation loss when using --reset-optimizer
Summary: Pull Request resolved: https://github.com/pytorch/fairseq/pull/661 Differential Revision: D15068312 Pulled By: myleott fbshipit-source-id: 1216835fd4c7f83ea5e350bff83901c93ac57447
0020477a · Myle Ott · Facebook Github Bot · fa52d202 · 0020477a
Commit 0020477a authored Apr 24, 2019 by Myle Ott Committed by Facebook Github Bot Apr 24, 2019
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

train.py train.py +1 -1

No files found.
--- a/train.py
+++ b/train.py
@@ -362,7 +362,7 @@ def load_checkpoint(args, trainer, epoch_itr):
            trainer.lr_step(epoch_itr.epoch)
            trainer.lr_step_update(trainer.get_num_updates())
-            if 'best' in extra_state:
+            if 'best' in extra_state and not args.reset_optimizer:
                save_checkpoint.best = extra_state['best']
        return True
    else: