Remove ds_config scheuduler params to prevent deepseed from creating scheduler for ref_model #2224
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Outdated suggestions cannot be applied.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.
What does this PR do?
Fixes #2154
My opinion is that popping the scheduler params off the deepcopied ds_config is the best way to solve this. This prevents deepspeed from attempting to build a scheduler for the engine (which is fine because any returned scheduler is ignored anyways). I explored some other avenues like using the Deepspeed inference engine instead but the different config would make it a nightmare.
Other:
ref_model
is None checked before_prepare_deepspeed
is called:also even if
_prepare_deepspeed
ends up getting used elsewhere passingmodel=None
to deepspeed will hit an assert anyways._prepare_deepspeed
can only be used on reference models that won't be trainedBefore submitting
Pull Request section?
to it if that's the case.
documentation guidelines.