Load models in-place #32
Merged
Conversation
This file contains 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
Before this change, we would first construct a model from using the arguments passed to the registry function. Then we would construct it again using `from_hf_hub`. This was not only a performance issue, but also a correctness issues -- the model constructed through `from_hf_hub` could have different hyperparameters than those specified in the arguments to the registry function. This change fixed this by using the new in-place loading support in Curated Transformers 2.0. The addition to in-place loading also added the `dtype` argument to the model configuration. We also expose this argument now in v2 versions of the registry functions. The configuration filling is also updated to fill the data type from the `torch_dtype` option in the HF model configuration.
danieldk
force-pushed
the
feature/inplace
branch
from
8e58ebd
to
fa27ea6
Compare
Otherwise fails on GPU, but the difference is too small to make a difference.
|
GPU test run: ✅ |
shadeMe
approved these changes
Apr 10, 2024
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Outdated suggestions cannot be applied.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.
Description
Before this change, we would first construct a model from using the arguments passed to the registry function. Then we would construct it again using
from_hf_hub. This was not only a performance issue, but also a correctness issues -- the model constructed throughfrom_hf_hubcould have different hyperparameters than those specified in the arguments to the registry function.This change fixed this by using the new in-place loading support in Curated Transformers 2.0.
The addition to in-place loading also added the
dtypeargument to the model configuration. We also expose this argument now in v2 versions of the registry functions. The configuration filling is also updated to fill the data type from thetorch_dtypeoption in the HF model configuration.Types of change
Bugfix/feature
Checklist