[he-yufeng]

Summary

Gemma3nAudioInputs declares input_features_padded and input_features_mask with fixed sequence-length dimensions. When two concurrent requests carry audio clips of different durations, TensorSchema.validate() sees mismatched shapes and kills EngineCore:

ValueError: input_features_padded contains inconsistent shapes:
  torch.Size([496, 128]) (index 0) vs torch.Size([449, 128]) (index 1)

The fix adds dynamic_dims={"s"} to both fields, matching the pattern established by PR #31223 for MiniCPM-o, Qwen2Audio, and AudioFlamingo3.

Changes

• vllm/model_executor/models/gemma3n_mm.py: Add dynamic_dims={"s"} to input_features_padded and input_features_mask in Gemma3nAudioInputs

Test Plan

• Verify concurrent audio requests to Gemma3n no longer crash
• Existing multimodal tests pass

Fixes #38297

[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.

[gemini-code-assist]

Code Review

This pull request updates the Gemma3nAudioInputs schema in vllm/model_executor/models/gemma3n_mm.py to explicitly mark the sequence dimension 's' as dynamic for both input_features_padded and input_features_mask. I have no feedback to provide.

[he-yufeng]

bump — 2-line fix, same pattern as #31223.

[ereish64]

I was having the same issue. Thanks for the PR.

[he-yufeng]

Rebased this onto current main to rerun the stale pre-run gate. The previous failure was from 2026-03-30 and required either a ready label or 4+ merged PRs; the author now has 5 merged vLLM PRs, so the gate should pass on the refreshed run. Local checks: py_compile on vllm/model_executor/models/gemma3n_mm.py and git diff --check.

[he-yufeng]

Rebased this branch onto current main and re-ran the lightweight checks for the Gemma3n audio schema change.

Validation:

python -m ruff check vllm\model_executor\models\gemma3n_mm.py
python -m ruff format --check vllm\model_executor\models\gemma3n_mm.py
python -m py_compile vllm\model_executor\models\gemma3n_mm.py
git diff --check

[he-yufeng]

Closing this during PR backlog cleanup. This one has not received human maintainer signal and is lower priority than the remaining vLLM PRs that are approved, actively reviewed, or more recent/high-signal. I can reopen or refresh it if maintainers want this specific fix.