Add balanced strategies for device_map in from_pretrained #18349
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
|
The documentation is not available anymore as the PR was closed or merged. |
LysandreJik
approved these changes
Aug 1, 2022
Comment on lines
+88
to
+91
| if version.parse(accelerate_version) > version.parse("0.11.0"): | ||
| from accelerate.utils import get_balanced_memory | ||
| else: | ||
| get_balanced_memory = None |
Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>
barry-jin
pushed a commit
to barry-jin/transformers
that referenced
this pull request
Sep 8, 2022
…e#18349) * Add balanced strategies for device_map in from_pretrained * Add safeguards for Accelerate version * Update src/transformers/modeling_utils.py Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr> * Style Co-authored-by: Lysandre Debut <lysandre.debut@reseau.eseo.fr>
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Outdated suggestions cannot be applied.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.
What does this PR do?
This PR brings to Transformers the functionality introduced in huggingface/accelerate#534 .
Basically
device_mapcan now take several options:"sequential"which corresponds to the current auto: fill each GPU sequentially (and if the user has lots of GPU spaces, some are not used at all)"balanced"which will split the model evenly across GPUs"balanced_low_0"which will split the model evenly across GPUs while leaving the most available memory on GPU 0, since that GPU might have more tensors on it when the outputs are used for some form of post-processing (generate and use_cache for instance)"auto"which now defaults to"balanced".When the user does not have enough GPU memory to accommodate the model, all the options are equivalent.