[AndreasKaratzas]

• registry.py: cap the SmolVLM HF reference at transformers<=5.19 with an hf reason, so test_common.py skips the HF-vs-vLLM comparison while vLLM-side tests keep running.

#61983 bumped Transformers to 5.20.2, and since then Multimodal Models (Extended Generation 2) shards 1 and 4 fail in build 96024. The HF reference SmolVLMModel.inputs_merger now splits image features into blocks of vision_config.patch_size (14) instead of the image token count (81 for SmolVLM2-2.2B), so the HF forward raises shape '[-1, 14, 2048]' is invalid; vLLM's SmolVLM is not involved. The Qwen2-VL / Qwen2.5-VL failure in each shard is the next test starting while the failed HF model still holds GPU memory (27.4 of 32.5 GiB free vs 29.9 GiB needed), and the test after it gets the memory back. huggingface/transformers#49725 fixed it on Transformers main on Oct 19, after 5.20.2 was cut, so the bound can go when vLLM moves to a release that includes it.

On MI355 with Transformers 5.20.2, pytest models/multimodal/generation/test_common.py on the two SmolVLM cases and the two Qwen cases that follow them in CI order: without the change the SmolVLM case fails the same way; with it, both SmolVLM cases skip and both Qwen cases pass in 12 runs. Five other runs failed only because another job on the shared node took up to 88 GB of the GPU.

Not a duplicate: no open PR touches SmolVLM or this failure.

AI assistance (Claude Code) was used to draft this change. Every line was reviewed by the submitter.

---

Pull Request Checklist

• I used vLLM's /pr-checklist skill. (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.

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[AndreasKaratzas]

/ci run

[github-actions]

✅ Triggered Buildkite CI #96077 for commit 0691f1322329.