Add optional support for Unsloth #1657
Draft
+756
−9
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
georgedouzas
force-pushed
the
georgedouzas/unsloth-fine-tuning-support
branch
from
f4832e3 to
acf918d
Compare
georgedouzas
force-pushed
the
georgedouzas/unsloth-fine-tuning-support
branch
from
acf918d to
7788ed7
Compare
georgedouzas
force-pushed
the
georgedouzas/unsloth-fine-tuning-support
branch
from
7788ed7 to
a34a3ea
Compare
oelachqar
reviewed
Nov 13, 2025
|
|
||
| from oumi.utils import logging | ||
|
|
||
|
|
Contributor
There was a problem hiding this comment.
Do we need this import here?
| peft_params=config.peft if use_peft else None, | ||
| **(additional_model_kwargs or {}), | ||
| ) | ||
| if is_unsloth_model(config.model.model_name): |
Contributor
There was a problem hiding this comment.
Given build_model already handles unsloth models, do we need this check here?
| "UP035", # UP035 `typing.List` is deprecated, use `list` instead | ||
| "UP006", # UP006 Use `list` instead of `List` for type annotation | ||
| ] | ||
| "src/oumi/__init__.py" = [ |
Contributor
There was a problem hiding this comment.
Is this change required?
| tokenizer: The tokenizer object built from the configuration. | ||
| """ | ||
| # Unsloth | ||
| if is_unsloth_model(model_params.model_name): |
Contributor
There was a problem hiding this comment.
Can we just use the regular tokenizer?
Contributor
|
@georgedouzas thank you for the contribution! At a high level, the approach looks great. If possible, it would be great to:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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
This PR adds minimal support for the unsloth library without making it a required dependency of Oumi. The idea is to let users load models and tokenizers through Unsloth’s API if it’s available, without changing the trainer abstraction or adding a new TrainerType. Since Unsloth patches
transformersandtrlunder the hood, it can work with the existing setup without needing a brand-new trainer.Feedback on whether this approach makes sense would be appreciated, or let me know if you would prefer a more explicit integration — like introducing a dedicated trainer type for Unsloth even though it’s patching everything internally. There are also some dependency conflicts between Unsloth and Oumi at the moment.
Related issues
#1322
Before submitting
Reviewers
At least one review from a member of
oumi-ai/oumi-staffis required.