New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Introduce logging_strategy training argument #10267
Introduce logging_strategy training argument #10267
Conversation
d70feae
to
d89b14f
Compare
Currently WIP. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This is a useful addition but it needs to be backward compatible: this means the default should give the same behavior as before (which is not the case right now as highlighted by the failing tests) so "steps" for the default logging_strategy and keep the eval_steps
default untouched either.
Could you amend your PR in that direction? Thanks!
d89b14f
to
6e9a757
Compare
Thanks, yes that worked out! |
It will just be ignored in that case, so there is no weird behavior for the user. |
Sure! That makes sense. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM! I'm just wondering is we should add the LoggingStrategy.NO
in case a user doesn't want any log since it comes for free (I don't think you'll need to change any line of code apart adding this to the enum).
This is because the next step is to do the same thing for saves (if you want to tackle it in a next PR!) which also would have the values steps/epochs/no so we could rename EvaluationStrategy
to something more generic like TimeInterval
(or any better name) which we would use for the evaluation strategy, saving strategy and logging strategy.
Introduce logging_strategy training argument in TrainingArguments and TFTrainingArguments. (huggingface#9838)
6e9a757
to
b3ed60c
Compare
Yes, something like |
Great! We can merge this in the meantime. |
Introduce logging_strategy training argument
in TrainingArguments and TFTrainingArguments. (#9838)
What does this PR do?
logging_strategy
argument in TrainingArguments.EvalStrategy
.Fixes #9838
Before submitting
Pull Request section?
to it if that's the case. Link to issue.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors which may be interested in your PR.
As changes in trainer: @sgugger