Fix empty LR scheduler for old jobs, clean up argument validation#288
Merged
Fix empty LR scheduler for old jobs, clean up argument validation#288
Conversation
mryab
commented
Mar 31, 2025
src/together/resources/finetune.py
Outdated
Member
Author
There was a problem hiding this comment.
As you can see here, we passed lr_scheduler_args despite this field not being defined for FinetuneCosineLRScheduler
Contributor
There was a problem hiding this comment.
Oops ty for the catch. I'm actually not sure it didn't have a validation error before...
src/together/types/finetune.py
Outdated
There was a problem hiding this comment.
Smart! Looks good to me if we have no other options to support old jobs
thepowerfuldeez
approved these changes
Apr 1, 2025
azahed98
approved these changes
Apr 1, 2025
70be7ec to
eeafb4a
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The introduction of
Literalin #273 broke parsing of old jobs without the learning rate scheduler specified, as Pydantic receives an empty string for the type and can't match it to any of the LR scheduler types that we have defined.The error looks like this:
As a simple fix before any backend changes, this PR adds an empty scheduler type and enables it only in
FinetuneResponse.together fine-tuning listworks with the new fixThe PR also makes a few cleanups around the codebase related to LR scheduling and fixes a hidden bug introduced in the previous PR, where
lr_scheduler_argswas renamed tolr_schedulerbut Pydantic silently accepted the previous name of the field