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Outdated suggestions cannot be applied.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.
Description
Currently, the HF estimator expects a HF model that is on the cpu: if a user moves the model to the GPU before giving it to the estimator there will be a crash to to a mismatch between the tensors created for the forward hooks in the HF model wrapper and the model. This small update ensures that the correct device is used between model and hook inputs.
Fixes # TO CREATE
Type of change
Please check all relevant options.
Testing
As the CI tests are CPU limited we have tested this via updating the notebook to shift the models to GPU devices in a few places before feeding to the estimator.
Test Configuration:
Checklist