[https://nvbugs/6423991][fix] Qwen3.5-VL MoE DFlash compatibility

[amukkara]

@coderabbitai summary

Description

DFlash failed to load on Qwen3.5-35B-A3B with AttributeError: 'Qwen3_5MoeVLModel' object has no attribute 'draft_config'.

model loader looks up draft_config/draft_model/load_draft_weights on the top-level VLM wrapper while they live on the inner SpecDecOneEngineForCausalLM. This PR forwards those three attributes from Qwen3VLModelBase wrapper to self.llm. Qwen3.5-VL MoE was added in #14599.

Test Coverage

TestQwen3_5_35B_A3B::test_fp8_moe_dflash

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

To see a list of available CI bot commands, please comment /bot help.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds speculative decoding forwarding to Qwen3VLModelBase by introducing draft_config and draft_model properties and a load_draft_weights method that delegate to the wrapped inner causal LM instance.

Changes

Draft API Forwarding

Layer / File(s)	Summary
Forward draft APIs to inner LM tensorrt_llm/_torch/models/modeling_qwen3vl.py	Adds draft_config and draft_model properties and a load_draft_weights method to Qwen3VLModelBase that delegate to self.llm.

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

Related Issues: None specified.

Related PRs: None specified.

Suggested labels: None specified.

Suggested reviewers: None specified.

Poem: A rabbit hops through code so neat, forwarding drafts, a tidy feat. Config and model, weights in tow, delegated onward, watch them flow. 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title matches the change and follows the required ticket/type format.
Description check	✅ Passed	The description includes the issue, solution, test coverage, and checklist, so it matches the template well.
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.}

[coderabbitai]

🧹 Nitpick comments (1)

tensorrt_llm/_torch/models/modeling_qwen3vl.py (1)

1309-1321: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

LGTM — forwarding matches the established pattern.

The forwarding here mirrors load_draft_weights in modeling_mistral.py, correctly resolving the AttributeError on draft_config/draft_model/load_draft_weights described in the PR.

One nit: none of the three new members has a return type annotation, unlike neighboring properties in this class (mm_token_ids, vocab_size_padded).

♻️ Suggested annotation additions

     `@property`
-    def draft_config(self):
+    def draft_config(self) -> Any:
         return getattr(self.llm, "draft_config", None)

     `@property`
-    def draft_model(self):
+    def draft_model(self) -> Any:
         return getattr(self.llm, "draft_model", None)

-    def load_draft_weights(self, *args, **kwargs):
+    def load_draft_weights(self, *args, **kwargs) -> None:
         return self.llm.load_draft_weights(*args, **kwargs)

As per path instructions, "Always annotate functions with return types (use None if no return); annotate class members and variables when necessary, especially for dataclasses and NamedTuple."

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tensorrt_llm/_torch/models/modeling_qwen3vl.py` around lines 1309 - 1321, The
new forwarding members in Qwen3VLModel are missing explicit return type
annotations, unlike neighboring properties. Add return annotations to
draft_config and draft_model in the Qwen3VL class, and mark load_draft_weights
as returning None while preserving the existing delegation to self.llm. Use the
established pattern from nearby properties like mm_token_ids and
vocab_size_padded to keep the API consistent.

Source: Path instructions

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@tensorrt_llm/_torch/models/modeling_qwen3vl.py`:
- Around line 1309-1321: The new forwarding members in Qwen3VLModel are missing
explicit return type annotations, unlike neighboring properties. Add return
annotations to draft_config and draft_model in the Qwen3VL class, and mark
load_draft_weights as returning None while preserving the existing delegation to
self.llm. Use the established pattern from nearby properties like mm_token_ids
and vocab_size_padded to keep the API consistent.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 34f358cc-cd32-4cec-8302-b9e65f28a509

📥 Commits

Reviewing files that changed from the base of the PR and between 9208094 and 627596b.

📒 Files selected for processing (1)

• tensorrt_llm/_torch/models/modeling_qwen3vl.py

[moraxu]

The same change will be merged as part of #15249 - feel free to close?

[amukkara]

#15249

sure, I will close this since #15249 is close to merging.