Upgrade transformers to >=5.0.0#169
Merged
LxYuan0420 merged 1 commit intovllm-project:mainfrom Mar 18, 2026
Merged
Conversation
Signed-off-by: Chao-Ju Chen <ricky.chen@infinirc.com>
LxYuan0420
approved these changes
Mar 18, 2026
Collaborator
LxYuan0420
left a comment
LxYuan0420
left a comment
There was a problem hiding this comment.
Thanks Ricky. I’m OK merging this as a temporary bridge.
Please follow up with a separate PR to bump mlx-lm / mlx-vlm minimums (and ideally a quick smoke run on a "needs Transformers v5" model) so the benefit is explicit. Once upstream vLLM lands the official v5 upgrade, we should drop the override in install.sh.
Contributor
Author
I opened: #174 (bump mlx-lm/mlx-vlm + Qwen3.5-0.8B smoke test) |
LxYuan0420
pushed a commit
that referenced
this pull request
Mar 19, 2026
## Summary Follow-up to #169 (transformers>=5.0.0 upgrade). - Bump `mlx>=0.31.0`, `mlx-lm>=0.31.0`, `mlx-vlm>=0.4.0` - Add `tests/test_qwen35_smoke.py`: golden token comparison for Qwen/Qwen3.5-0.8B (greedy decoding, 5/5 passed) Qwen3.5 uses the `qwen3_5` architecture which requires transformers>=5.0.0 and mlx-lm>=0.30.0. This proves the upgraded dependency stack works end-to-end on Metal. ## Test ### test_qwen35_smoke <img width="1087" height="628" alt="截圖 2026-03-18 下午5 09 56" src="https://github.com/user-attachments/assets/f58f4883-1776-4f69-b2d1-10a1bad2938d" /> ### test_paged_deterministic <img width="1108" height="637" alt="截圖 2026-03-18 下午5 10 22" src="https://github.com/user-attachments/assets/f501d7d0-714c-4b05-931f-00ce55444240" /> Signed-off-by: Chao-Ju Chen <ricky.chen@infinirc.com>
This was referenced Mar 19, 2026
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
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.
Summary
Upgrade transformers from >=4.40.0 to >=5.0.0. mlx-lm 0.30+ and mlx-vlm 0.3.10+ require transformers 5.x for newer model architectures (Qwen3.5,Nemotron, etc.). vllm upstream is tracking the official upgrade in vllm-project/vllm#30566 but it hasn't landed yet — this unblocks the MLX side first.
Test results
All tests run with vllm 0.17.1 + transformers 5.3.0:
scripts/test.sh): server starts, chat completions passtest_paged_deterministic.py):