[#16792][fix] Preserve Qwen3.5-VL multimodal content order

[fallintoplace]

Description

Qwen3.5-VL uses string chat content, but its registration did not opt into the serving path that preserves ordered OpenAI content parts. Requests containing alternating text and images therefore bulk-prepended all vision placeholders, changing prompt tokenization and multimodal positions.

Enable the existing interleaved-placeholder path for the shared dense and MoE metadata. Requests without ordered content parts keep the existing BEFORE_TEXT fallback.

This addresses the placeholder-ordering part of #16792. The separate residual generation parity case described there remains out of scope.

Test Coverage

• Extended the dense and MoE placeholder-registration tests to cover the interleaving contract.
• pre-commit run --files tensorrt_llm/_torch/models/modeling_qwen3_5.py tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py
• python3 -m compileall -q tensorrt_llm/_torch/models/modeling_qwen3_5.py tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py

PR Checklist

• Reviewed the checklist in the pull request template as applicable.

Dev Engineer Review

• Enabled interleaved placeholder registration for shared Qwen3.5-VL dense and MoE metadata.
• Configuration is consistent with the intended OpenAI content-part ordering behavior and preserves the existing fallback path.
• Added metadata assertions for both variants. No apparent API, performance, or error-handling regressions.
• Residual generation parity remains out of scope.

QA Engineer Review

• Modified placeholder-metadata tests for qwen3_5 and qwen3_5_moe, adding assertions that interleave_placeholders is enabled.
• No corresponding test-db/ or qa/ coverage entries were identified.
• Verdict: needs follow-up — CI/manual coverage data is unavailable.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: ad2b34e9-e38c-4ef6-91c9-79994b136666

📥 Commits

Reviewing files that changed from the base of the PR and between f5c2a07 and 50864ce.

📒 Files selected for processing (3)

• tensorrt_llm/_torch/models/modeling_qwen3_5.py
• tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py
• tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py

---

Walkthrough

Qwen3.5 VLM placeholder metadata now enables interleaving for multimodal inputs. Registration tests for the dense and MoE variants verify the new metadata setting.

Changes

Qwen3.5 placeholder metadata

Layer / File(s)	Summary
Enable and verify placeholder interleaving tensorrt_llm/_torch/models/modeling_qwen3_5.py, tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl.py, tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py	Shared Qwen3.5 VLM metadata sets interleave_placeholders=True, with registration tests covering the dense and MoE configurations.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Suggested reviewers: qijune

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the PR's main change and follows the required [issue][type] summary format.
Description check	✅ Passed	The description includes the issue, solution, test coverage, and PR checklist sections required by the template.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}