[None][feat] Revert Add support for Qwen3.5 VL MoE (#14164)

[nv-guomingz]

This reverts commit 96a4a09.

Summary by CodeRabbit

• Documentation

  • Removed Qwen3.5 MoE Vision Language model from the supported models matrix.

• Refactor

  • Simplified model weight loading and configuration handling for supported Qwen3.5 variants.
  • Consolidated configuration normalization logic for improved maintainability.

• Tests

  • Removed test coverage for the discontinued Qwen3.5 MoE Vision Language model variant.
  • Updated related test configurations and lists accordingly.

Description

Test Coverage

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.

[nv-guomingz]

/bot run --disable-fail-fast

[nv-guomingz]

The #14164 blocked the L0, please refer to #14455

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR removes Qwen3.5-MoE VL conditional generation model support by deleting model exports, updating weight mapper registration, relocating config normalization logic, simplifying model signatures and KV-cache handling, and removing all related tests and documentation.

Changes

Qwen3.5-MoE-VL Support Removal and Config Refactoring

Layer / File(s)	Summary
Model export and documentation cleanup tensorrt_llm/_torch/models/__init__.py, docs/source/models/supported-models.md	Remove Qwen3_5MoeVLModel from public module exports and __all__ registry; delete conditional generation model from supported models documentation.
Weight mapper registration update tensorrt_llm/_torch/models/checkpoints/hf/qwen3_5_weight_mapper.py	Reregister Qwen3_5MoeHfWeightMapper to Qwen3_5MoeForCausalLM instead of Qwen3_5MoeForConditionalGeneration.
Config compatibility consolidation and relocation tensorrt_llm/_torch/models/modeling_qwen3_5.py, tensorrt_llm/_torch/pyexecutor/config_utils.py	Move Qwen35ConfigCompat from modeling to config utilities as _Qwen35ConfigCompat, consolidating text-config extraction, architecture selection, quantization module rewriting, linear-attention bf16 workaround, and RoPE parameter flattening; update modeling imports and docstrings.
Config loading normalization tensorrt_llm/_torch/pyexecutor/config_utils.py	Update load_pretrained_config to use local _Qwen35ConfigCompat.normalize() for Qwen3.5 configs and remove older VLM normalization branch.
KV cache dtype resolution simplification tensorrt_llm/_torch/pyexecutor/config_utils.py, tensorrt_llm/_torch/pyexecutor/model_loader.py	Remove dtype-coercion helpers and update mamba KV cache dtype resolution to use direct config field access, eliminating fallback chains.
Model signature simplification tensorrt_llm/_torch/models/modeling_qwen3_next.py	Simplify Qwen3NextForCausalLM.load_weights to accept only weights and weight_mapper, removing optional params_map and allow_partial_loading parameters.
VL model architecture support removal tensorrt_llm/_torch/models/modeling_qwen3vl.py	Remove conditional generation architecture branch and simplify head dimension calculation in Qwen3VLModelBase.
Test utility simplification tests/unittest/_torch/modeling/test_modeling_multimodal.py	Remove hybrid KV-cache manager support and related hooks; simplify init_kv_cache_manager to always use standard get_kv_cache_manager path; clean up related imports and setUp logic.
Test case and configuration removal tests/integration/defs/accuracy/references/mmmu.yaml, tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py, tests/integration/test_lists/qa/llm_function_core.txt, tests/integration/test_lists/test-db/l0_l40s.yml, tests/unittest/_torch/modeling/test_modeling_qwen3_5_vl_moe.py	Remove TestQwen3_5_35B_A3B_VL test class and test entries; update MMMU config to explicit bfloat16 dtype; delete entire Qwen3.5 MoE-VL test module.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#14164: This PR reverses the Qwen3.5 MoE-VL support addition by removing the model exports, weight mapper registration, and test coverage that were previously introduced.

Suggested reviewers

• 2ez4bz
• liji-nv
• syuoni

🚥 Pre-merge checks | ✅ 3 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 54.55% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.
Description check	⚠️ Warning	The PR description is incomplete. While it identifies that this is a revert, it lacks an explanation of why the revert is necessary, what issues prompted it, and provides no test coverage information or completed PR checklist.	Add a clear rationale for the revert, describe the motivation and any related issues, list relevant test coverage that validates the revert, and complete the PR checklist items.

✅ Passed checks (3 passed)

Check name	Status	Explanation
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.
Title check	✅ Passed	The title clearly summarizes the main change: reverting support for Qwen3.5 VL MoE from the referenced commit, which matches the changeset's comprehensive removal of related code and tests.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch revert-14164-qwen3_5_vl_moe

---

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

[tensorrt-cicd]

PR_Github #49944 [ run ] triggered by Bot. Commit: ee593dc Link to invocation

[tensorrt-cicd]

PR_Github #49944 [ run ] completed with state SUCCESS. Commit: ee593dc
/LLM/main/L0_MergeRequest_PR pipeline #39515 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[moraxu]

/bot run

[tensorrt-cicd]

PR_Github #49997 [ run ] triggered by Bot. Commit: ee593dc Link to invocation

[tensorrt-cicd]

PR_Github #49997 [ run ] completed with state SUCCESS. Commit: ee593dc
/LLM/main/L0_MergeRequest_PR pipeline #39562 completed with status: 'SUCCESS'

CI Report

Link to invocation