Repository navigation
[CI] Bump Transformers version to 5.23.0 - #67502
Merged
Merged
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Member
Author
|
/ci run nightly |
|
✅ Triggered Buildkite CI #102497 for commit |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
DarkLight1337
approved these changes
Nov 13, 2026
Transformers v5.23 validates that placeholder counts match the passed media. Qwen3-VL, Gemma 4 and GLM-4.1V process videos separately, then called the HF processor with text still containing video placeholders. Rebuild that text without videos, and drop the code that expanded video tokens in the text or `input_ids`, since vLLM discards that output. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Without it, audio extracted from videos is counted as standalone audio and fails the placeholder count validation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Transformers v5.23 validates flat kwargs against the processor's kwargs `TypedDict`, so remote code that omits `total=False` (e.g. MiDashengLM) makes every kwarg required. Mark these as optional when the remote code is imported, and warn so the remote code can be fixed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
The remote `BailingMoeV3VL` processor's `__call__` is incompatible with the kwargs returned by `_merge_kwargs` in Transformers v5.23. Call its image processor directly, as is already done for videos. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Transformers v5.23 dropped `truncation: False` from the Qwen3-ASR audio defaults, so `WhisperFeatureExtractor` checkpoints truncate audio longer than 30 s in HF but not in vLLM. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
hmellor
requested review from
AndreasKaratzas,
vadiklyutiy and
ywang96
as code owners
November 13, 2026 19:07
Member
Author
|
/ci retry |
|
✅ The previous CI build is still running: https://buildkite.com/vllm/ci/builds/102497 |
mergify
Bot
added
multi-modality
Member
|
Could you run the full CI ( |
Member
Author
|
The initial run was nightly. I'm using retry to avoid being wasteful while burning down the errors. Then the last run when I think it'll all pass will be nightly too |
Member
Author
|
/ci retry |
|
✅ Triggered Buildkite CI #102660 for commit |
Member
Author
|
/ci retry |
|
✅ Queued 13 failed job(s) for retry in Buildkite CI #102660. |
6 tasks done
The remote code also imports BASE_VIDEO_PROCESSOR_DOCSTRING, which Transformers v5.15 removed, so the model cannot load on the vLLM side either. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Member
Author
|
/ci retry |
|
✅ Triggered Buildkite CI #102766 for commit |
Member
Author
|
/ci run nightly |
|
✅ Triggered Buildkite CI #102797 for commit |
Member
Author
|
/ci retry |
|
✅ Queued 2 failed job(s) for retry in Buildkite CI #102797. |
6 tasks done
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
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.
Overview
Bump Transformers to v5.23.0. Follows #64529.
Claims
requirements/common.txtallowstransformers >= 5.16.1, < 5.24.0, and CI pins 5.23.0.ep_routerremoval 🚨 Remove theep_routerexpert-parallel style huggingface/transformers#49615, exporters, Wav2Vec2ProcessorWithLM, serve) do not touch vLLM._get_hf_mm_inputs), so this should only hit models that override that path.LlamaForCausalLM, and the native InternLM3 config exposesattention_bias/mlp_bias, whichllama.pyreads.5.23.0.dev0([Model] Add transcription to the Transformers modeling backend #65735) now run in CI.Validation
The HF PyPI mirror does not have 5.23.0 yet, so the compiled requirements were edited by hand (pin and hashes from pypi.org, committed with
SKIP=pip-compile) and this is validated by/ci run nightly. Results to follow.Pull Request Checklist
I used vLLM's
/pr-checklistskill. (Mandatory for agents, optional for humans).AI assistance was used during the creation of this PR.
Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.
Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.
Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.
Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.
🤖 Generated with Claude Code