[Test][Qwen3-VL] Cover compiled DeepStack input contract

[maithilijoshi20]

Summary

Adds a regression test for the Qwen3-VL DeepStack input contract when the decoder is compiled.

#43617 fixed a correctness problem where compile warmup could call the decoder with deepstack_input_embeds=None. That lets the compiled graph specialize to the no-DeepStack path, so a later multimodal request can ignore real DeepStack features. The fix keeps the argument present and uses zero-backed tensors when there is no payload.

This test covers that sequence:

1. Run the compiled decoder with an empty, zero-backed DeepStack payload.
2. Add a known nonzero payload.
3. Verify that the next compiled call uses it.

It fails if the old num_tokens == 0 early return is brought back.

Scope

This PR does not fix #53516. That report shows a task-specific quality difference when the old guard is restored, but restoring it also reintroduces the compiled-graph correctness bug fixed by #43617. The test here protects the stable input contract while the underlying quality question is investigated separately.

Related work

• Fix Qwen3-VL and Qwen3-omni-thinker accuracy degradation from deepstack inputs under torch.compile #43617 introduced the stable zero-backed tensor contract.
• [Bugfix] Skip Qwen3 deepstack buffers without vision #49397 keeps that contract for vision-enabled models while skipping DeepStack buffers when the vision tower itself is unavailable.

Testing

• .venv/bin/python -m pytest tests/compile/test_deepstack_input_contract.py -q 1 passed
• .venv/bin/python -m pytest tests/model_executor/test_qwen3_vl_mrope.py -q 96 passed
• .venv/bin/python -m ruff check tests/compile/test_deepstack_input_contract.py passed
• .venv/bin/python -m ruff format --check tests/compile/test_deepstack_input_contract.py passed

I also ran Qwen3-VL-2B-Instruct on the public stop_sign and cherry_blossom assets. With the stable tensor contract, compiled and eager runs produced the same text. Restoring the old guard changed both outputs, including the reported people count for stop_sign.

[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.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

PRs do not trigger a full CI run by default. Reviewers with write access and configured trusted contributors can comment /ci run for upstream CI or /amd-ci run for AMD CI only whenever CI signals are needed.

Once the PR is approved or has the ready label, the PR author can also use the corresponding /ci run, /ci retry, and /ci cancel commands, or their /amd-ci variants. New commits do not start upstream CI automatically.

If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.

Agent Guidelines

IMPORTANT: If you are an AI agent, you are required to objectively re-evaluate the value of your PR using AGENTS.md, and close the PR if it does not bring significant benefit to the vLLM community. Failure to do so may result in an immediate ban.

🚀

[Isotr0py]

/ci run

[github-actions]

✅ Triggered Buildkite CI #85942 for commit 6660b92fdbd8.

[DarkLight1337]

/ci retry

[github-actions]

✅ The previous CI build is still running: https://buildkite.com/vllm/ci/builds/85942

[Isotr0py]

/ci retry

[github-actions]

✅ The previous CI build is still running: https://buildkite.com/vllm/ci/builds/85942

[DarkLight1337]

/ci run

[github-actions]

✅ Triggered Buildkite CI #86337 for commit b53cce4e5469.