Update transformers to 4.51 #1790
Merged
Conversation
This file contains hidden or 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
ethantang-db
approved these changes
Apr 16, 2025
irenedea
approved these changes
Apr 16, 2025
dakinggg
added a commit
to databricks/compose-rl
that referenced
this pull request
Apr 17, 2025
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Outdated suggestions cannot be applied.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.
In the latest transformers version, they added a py.typed file, which means that the type checker starts using types from transformers. There are a number of legitimitate type issues surfaced that I tried to correct, but the vast majority result from transformers not actually being a type checked library and a couple specific difficulties:
(1) BatchEncoding (which is what tokenizers return) has a getitem whose return type depends on the type of the index passed in (string vs int). The type checker can't handle this (correct answer would be to have overloads I believe).
(2) PeftModel forwards attr access to the underlying PreTrainedModel by overriding getattr. The type checker can't know this. The correct answer would be to do lots of unpacking of the model before accessing attrs, but I don't think its worth the energy.
Generally, if it wasn't clear to me what to do, I default to type ignore, since that is no worse of a state than before. For reviewers, probably best to skip reviewing all the type ignores, and focus on any places where I actually changed code.
Manual test run:
tr-types-2-9qAwxu