[torchax]: JittableModule statedict handling #9195
Open
+137
−0
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Outdated suggestions cannot be applied.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.
torchax aims to improve seamless interoperability between torch and jax
one of the parts in torch training pipeline revolves around storing and loading statedict (checkpoints)
most of the objects revolving torch checkpoints expect a (non nested) dict containing weight name and it's value (in either CPU or GPU device)
since torchax tensors are held in jax container, torch checkpointers cannot easily handle it
this changes forces JittableModule to convert state_dict() functions (both load and get), making it seamless to the user when he wants to extract the statdict prior to saving it as a checkpoint