[TRTLLM-14609][chore] Remove legacy MoE path in DenseGEMMFusedMoE

[xxi-nv]

Description

For the sake of MoE code health, this PR removes the legacy (pre-ConfigurableMoE) forward path from DenseGEMMFusedMoE.

ConfigurableMoE has been the default MoE path for months (ENABLE_CONFIGURABLE_MOE defaults to "1"), so the backend-internal forward pipeline is no longer reachable in the default flow.

What is removed (tensorrt_llm/_torch/modules/fused_moe/fused_moe_densegemm.py)

• forward_chunk()
• forward_impl()

Net effect: 75 deletions in a single file.

What is deliberately preserved

• The live contract called by ConfigurableMoE / MoEScheduler is untouched: run_moe() and supports_moe_output_in_alltoall_workspace().

Test Coverage

Verified on OCI GB200 (1 node, 4 GPUs) with the container pinned by jenkins/current_image_tags.properties, running tests/unittest/_torch/modules/moe/test_moe_module.py filtered to this backend:

Scope	Result
test_configurable_moe_single_gpu	5 passed
test_configurable_moe_multi_gpu	8 skipped (pre-existing suite skips, see note)
test_configurable_moe_multi_gpu_eplb	no matching cases for this backend

5 passed, 0 failed.

Note on multi-GPU coverage. All 8 collected multi-GPU cases for DENSEGEMM (parallel=DTP/TTP x comm=NVLINK_ONE_SIDED/TWO_SIDED x quant=NVFP4) are skipped by the existing skip logic in tests/unittest/_torch/modules/moe/moe_test_utils.py. That file is not touched by this PR, so these cases skip identically on unmodified main — the skips are pre-existing and not caused by this change. The two removed methods are not referenced by ConfigurableMoE or MoEScheduler, so the residual risk is low, but reviewers should be aware that multi-GPU coverage for this backend comes from the unchanged shared scheduler path rather than from backend-specific cases.

PR Checklist

• PR title follows [JIRA/NVBUG][type] description
• Commit is signed off (DCO)
• Tests pass for the affected backend

Dev Engineer Review

• Removed the unused legacy forward_chunk() and forward_impl() methods, along with the obsolete allgather import, from DenseGEMMFusedMoE.
• The active ConfigurableMoE interface remains unchanged, including run_moe() and supports_moe_output_in_alltoall_workspace().
• No configuration or test-list files were modified.
• Reported validation: 5 single-GPU backend tests pass; 8 multi-GPU cases remain skipped by existing test logic.

QA Engineer Review

No test changes.

[xxi-nv]

/bot run

[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: 5096415b-a82e-4096-9e7b-fe746f2985c3

📥 Commits

Reviewing files that changed from the base of the PR and between 4b7d719 and 43cee82.

📒 Files selected for processing (1)

• tensorrt_llm/_torch/modules/fused_moe/fused_moe_densegemm.py

💤 Files with no reviewable changes (1)

• tensorrt_llm/_torch/modules/fused_moe/fused_moe_densegemm.py

---

Walkthrough

The DenseGEMMFusedMoE implementation removes the unused allgather import and deletes the legacy forward_chunk and forward_impl methods, leaving the existing run_moe path intact.

Changes

DenseGEMM Fused MoE cleanup

Layer / File(s)	Summary
Remove legacy forward paths tensorrt_llm/_torch/modules/fused_moe/fused_moe_densegemm.py	Removes the unused import and deletes forward_chunk and forward_impl, including their routing, quantization, distributed communication, MoE execution, and reduction logic.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title follows the required ticket-and-type format and accurately summarizes removal of the legacy MoE path.
Description check	✅ Passed	The description includes the issue, solution, test coverage, and checklist, so it is mostly complete and on-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.}

[tensorrt-cicd]

PR_Github #61760 [ run ] triggered by Bot. Commit: 43cee82 Link to invocation

[tensorrt-cicd]

PR_Github #61760 [ run ] completed with state SUCCESS. Commit: 43cee82
/LLM/main/L0_MergeRequest_PR pipeline #49962 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

[xxi-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #61820 [ run ] triggered by Bot. Commit: 43cee82 Link to invocation

[tensorrt-cicd]

PR_Github #61820 [ run ] completed with state FAILURE. Commit: 43cee82
/LLM/main/L0_MergeRequest_PR pipeline #50013 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

[xxi-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #62194 [ run ] triggered by Bot. Commit: 0fc1bb0 Link to invocation

[tensorrt-cicd]

PR_Github #62194 [ run ] completed with state FAILURE. Commit: 0fc1bb0
/LLM/main/L0_MergeRequest_PR pipeline #50365 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

[xxi-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #62264 [ run ] triggered by Bot. Commit: 0fc1bb0 Link to invocation

[tensorrt-cicd]

PR_Github #62264 [ run ] completed with state FAILURE. Commit: 0fc1bb0
/LLM/main/L0_MergeRequest_PR pipeline #50433 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

[xxi-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #62314 [ run ] triggered by Bot. Commit: 0fc1bb0 Link to invocation

[tensorrt-cicd]

PR_Github #62314 [ run ] completed with state FAILURE. Commit: 0fc1bb0
/LLM/main/L0_MergeRequest_PR pipeline #50484 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

[xxi-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #62342 [ run ] triggered by Bot. Commit: 0fc1bb0 Link to invocation

[tensorrt-cicd]

PR_Github #62342 [ run ] completed with state SUCCESS. Commit: 0fc1bb0
/LLM/main/L0_MergeRequest_PR pipeline #50510 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

CI Report

Link to invocation