Resetting of high scores, scheduler and optimizer for fine-tuning/domain adaptation #75
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In the case of fine-tuning or domain adaptation, one might want to overwrite the existing scheduler or optimizer or the previously tracked high score.
This is specifically relevant when fine-tuning with a different metric or a data where the validation score is in a different range than the loaded model (e.g. pre-training on BLEU and fine-tuning on PPL would result in never storing a new checkpoint). When the data is much smaller than the previous training data, one might want to reduce the patience.
We can also think about modeling this as "load_x_from_ckpt" rather than "reset_x", but for now, it seems that in the default case we want to load everything.