Bump transformers version to 4.48.1 #771
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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
In order to be able to be independent of optimum releases to update optimum-neuron transformer dependency, we now recommend a direct installation instead of installing optimum-neuron as an optimum extra.
dacorvo
force-pushed
the
bump_transformers_version
branch
from
668423c to
565639f
Compare
This is because otherwise contronel-aux will pull an extremely old version of timm.
The parallelization code expects these parameters to be set. A proper fix would be to write a specific Llama parallel model.
The transformation code expects some attributes to be set.
dacorvo
force-pushed
the
bump_transformers_version
branch
from
565639f to
2fc13fc
Compare
tengomucho
approved these changes
Jan 30, 2025
michaelbenayoun
approved these changes
Jan 30, 2025
Comment on lines
+590
to
+593
| # FIXME: temporary workaround to avoid too many changes in the transformation code | ||
| layer.self_attn.num_heads = layer.self_attn.config.num_attention_heads | ||
| layer.self_attn.num_key_value_heads = layer.self_attn.config.num_key_value_heads | ||
| layer.self_attn.hidden_size = layer.self_attn.config.hidden_size |
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Outdated suggestions cannot be applied.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 bumps the transformers version to 4.48.1 to allow the deployment of images containing
optimum-neuronin AWS services (all previous transformers versions are now rejected because they include some scripts identified as potentially harmful).This also modifies the documentation to change the installation instructions: installing as an optimum extra is no longer recommended and will be soon deprecated.
There were some changes in transformers modeling for LLama and Mistral models attention forward that needed to be reflected in the distributed parallelization modeling code.
There were also some changes in the attributes of the transformers Attention classes for the same models that broke the parallelization dynamic transformations: the changes in this pull-request are just a workaround, and eventually these dynamic transformations should be replaced by explicit modeling code (the forward methods are already overloaded, so there is no point in patching dynamically the layers where it would be simpler to recreate them directly from the config).