upgrade transformers==5.3.0 trl==0.29.0 kernels by winglian · Pull Request #3459 · axolotl-ai-cloud/axolotl

winglian · 2026-03-05T20:05:48Z

Description

Motivation and Context

How has this been tested?

AI Usage Disclaimer

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

Summary by CodeRabbit

Chores
- Updated dependencies including transformers, deepspeed, trl, hf_xet, and kernels to newer versions for improved compatibility and performance.
- Updated CI/CD pipeline CUDA configuration.
Tests
- Updated tokenizer test assertions and expected token sequences for improved accuracy.

coderabbitai · 2026-03-05T20:06:05Z

Important

Review skipped

Auto incremental reviews are disabled on this repository.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: fcdd8650-600b-4603-a3da-9f7677608fda

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

📝 Walkthrough

Walkthrough

This pull request updates dependencies in requirements.txt (transformers, deepspeed, trl, hf_xet, kernels), adjusts CUDA matrix entries in CI workflows from version 12.9.1 to 12.8.1, adds an optional model parameter to the OptimizerMixin.create_optimizer method, and updates test assertions for tokenizer type validation and expected token IDs.

Changes

Cohort / File(s)	Summary
CI Workflow Configuration `.github/workflows/tests.yml`	Replaced CUDA version 12.9.1 with 12.8.1 in docker-e2e-tests-1st and docker-e2e-cleanup job matrices.
Dependency Management `requirements.txt`	Updated multiple dependencies: transformers (5.2.0→5.3.0), trl (0.28.0→0.29.0), hf_xet (1.2.0→1.3.2), kernels (0.12.1→0.12.2), and adjusted deepspeed version constraint (>=0.18.3 to >=0.18.6,<0.19.0).
Core Trainer Modifications `src/axolotl/core/trainers/mixins/optimizer.py`	Added optional `model` parameter to `create_optimizer()` method; optimization model selection now respects provided parameter over instance model when specified.
Test Assertions `tests/prompt_strategies/test_chat_templates.py`, `tests/test_tokenizers.py`	Added LlamaTokenizer class assertion for phi35 tokenizer; updated expected token IDs in test_phi35 (EOT markers: 22172→12199; label sequences: 1781→16773) and test_tokenizers (added token path: [1, 32000, 1404]→[1, 32000, 1792]).

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

#3407: Updates transformers dependency version in requirements.txt with overlapping release pattern.
#3272: Modifies CI workflow CUDA matrix entries to include CUDA 12.8.1 configurations.
#3214: Updates both transformers and trl dependencies in requirements.txt as part of coordinated release.

Suggested labels

scheduled_release

Suggested reviewers

NanoCode012
SalmanMohammadi

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately reflects the main changes: upgrades to transformers (5.3.0), trl (0.29.0), and kernels, which are the primary dependency updates documented in requirements.txt.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests
Post copyable unit tests in a comment
Commit unit tests in branch transformers-530

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

github-actions · 2026-03-06T02:20:33Z

📖 Documentation Preview: https://69aabc1b2436da3456e58629--resonant-treacle-0fd729.netlify.app

Deployed on Netlify from commit bc23aa9

codecov · 2026-03-06T02:44:17Z

Codecov Report

❌ Patch coverage is 51.72414% with 14 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/axolotl/utils/schemas/deprecated.py	57.14%	6 Missing ⚠️
src/axolotl/core/trainers/dpo/trainer.py	0.00%	2 Missing ⚠️
src/axolotl/integrations/liger/plugin.py	60.00%	2 Missing ⚠️
src/axolotl/core/trainers/mixins/optimizer.py	66.66%	1 Missing ⚠️
src/axolotl/loaders/model.py	0.00%	1 Missing ⚠️
src/axolotl/loaders/tokenizer.py	0.00%	1 Missing ⚠️
src/axolotl/utils/data/shared.py	0.00%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

NanoCode012 · 2026-03-06T09:10:08Z

tests/core/test_builders.py

        [
            "sft_cfg",
-            "rm_cfg",
+            # "rm_cfg",  # TODO fix for num_labels = 2 vs 1


What does this mean?

NanoCode012 · 2026-03-06T12:05:43Z

tests/test_tokenizers.py

        tokenizer = load_tokenizer(cfg)
-        assert tokenizer("<|im_start|>user")["input_ids"] == [1, 32000, 1404]
+        assert "LlamaTokenizer" in tokenizer.__class__.__name__
+        assert tokenizer("<|im_start|>user")["input_ids"] == [1, 32000, 1792]


Why did this change?

there was a tokenizers change upstream

huggingface/transformers#43791

winglian added 4 commits March 5, 2026 15:05

upgrade transformers==5.3.0 trl==0.29.0 kernels

cc57ed0

use latest deepspeed fixes

c19be2b

use corect image for cleanup

e06f833

fix test outputs for tokenizer fixes upstream

c799c8d

winglian mentioned this pull request Mar 5, 2026

integration branch for transformers#44041 #3420

Closed

winglian added 11 commits March 5, 2026 15:23

fix import:

7054bce

keep trl at 0.28.0

f81f967

handle updated API

e5a5b82

use latest trl since 0.28.0 doesn't work with latest transformers

28c1f81

use trl experimental for pad to length

bdfbf72

monkeypatch trl with ORPOTrainer so liger doesn't croak

bc64ea0

upgrade accelerate

75aafb6

more fixes

3e0e319

move patch for orpotrainer

f45d188

load the imports later

83635d6

remove use_logits_to_keep

7e3de83

fix loss_type arg as a list

2c99c49

winglian requested a review from NanoCode012 March 6, 2026 02:42

winglian and others added 7 commits March 5, 2026 22:56

fetch hf cache from s3

14b7758

just manually download the missing model for now

5977e0d

lint for pre-commit update

75c5391

a few more missing models on disk

8155b1e

fix: loss_type internally now list

b19baf6

fix: remove deprecated code and raise deprecate

28cc4b9

fix: remove unneeded blocklist

b9dc87b

NanoCode012 reviewed Mar 6, 2026

View reviewed changes

tests/core/test_builders.py

[

"sft_cfg",

"rm_cfg",

# "rm_cfg", # TODO fix for num_labels = 2 vs 1

Copy link

Collaborator

NanoCode012 Mar 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What does this mean?

fix: remove reliance on transformers api to find package available

5bf5867

NanoCode012 added 2 commits March 6, 2026 18:17

chore: refactor shim for less sideeffect

4f183d3

fix: silent trl experimental warning

bc23aa9

NanoCode012 added the scheduled_release This PR is slated for the upcoming release label Mar 6, 2026

NanoCode012 reviewed Mar 6, 2026

View reviewed changes

winglian merged commit cada93c into main Mar 6, 2026
24 checks passed

winglian deleted the transformers-530 branch March 6, 2026 14:11

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

upgrade transformers==5.3.0 trl==0.29.0 kernels#3459

upgrade transformers==5.3.0 trl==0.29.0 kernels#3459
winglian merged 26 commits intomainfrom
transformers-530

winglian commented Mar 5, 2026 •

edited by coderabbitai bot

Loading

Uh oh!

coderabbitai bot commented Mar 5, 2026 •

edited

Loading

Review skipped

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Suggested labels

Suggested reviewers

❌ Failed checks (1 warning)

Uh oh!

github-actions bot commented Mar 6, 2026 •

edited

Loading

Uh oh!

codecov bot commented Mar 6, 2026 •

edited

Loading

Uh oh!

NanoCode012 Mar 6, 2026

Uh oh!

NanoCode012 Mar 6, 2026

Uh oh!

winglian Mar 6, 2026

Uh oh!

winglian Mar 6, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

winglian commented Mar 5, 2026 • edited by coderabbitai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Motivation and Context

How has this been tested?

AI Usage Disclaimer

Screenshots (if appropriate)

Types of changes

Social Handles (Optional)

Summary by CodeRabbit

Uh oh!

coderabbitai bot commented Mar 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Suggested labels

Suggested reviewers

❌ Failed checks (1 warning)

Uh oh!

github-actions bot commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov bot commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

NanoCode012 Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

NanoCode012 Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

winglian Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

winglian Mar 6, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

winglian commented Mar 5, 2026 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Mar 5, 2026 •

edited

Loading

github-actions bot commented Mar 6, 2026 •

edited

Loading

codecov bot commented Mar 6, 2026 •

edited

Loading