[#12290][fix] Qwen 3.5 fix 3d position ID handling

[bmarimuthu-nv]

Description

Fixes #12271

This updates the Qwen3.5 MoE AutoDeploy multimodal path to make mRoPE delta handling request-scoped and cache-backed, while keeping Qwen-specific multimodal span construction in the Qwen model path instead of the generic input processor.

Implemented changes:

• derive exact Qwen multimodal spans from the tokenized input_ids in the Qwen-specific AutoDeploy path
• build chunk-aware multimodal 3D positions in the VLM wrapper
• add request-level mRoPE delta helpers and wire the wrapper to use a per-request mrope_delta_cache when available
• add the initialize_qwen_mrope_delta_cache transform to allocate a tiny per-slot cache resource
• make metadata-free prefill warmups a zero-delta no-op for the cached op
• fix the flattened cached position_ids fallback so request-level deltas are expanded to per-token deltas using cu_seqlen
• remove the temporary reduced-layer cap in the Qwen3.5 35B MoE AutoDeploy accuracy test after the e2e path was revalidated

Test Coverage

Verified locally:

• pytest -q tests/unittest/_torch/auto_deploy/unit/singlegpu/models/test_qwen3_5_moe.py -k "mrope_delta_matches_compute_mrope_positions_for_mixed_items or mrope_delta_with_cache_writes_prefill_and_reads_decode or flattened_prefill_with_request_deltas" -p no:cacheprovider
• TRTLLM_ACCURACY_NO_REFERENCE=1 TMPDIR=/tmp/trtqwen HF_HOME=/tmp/trtqwen/hf_home PYTHONDONTWRITEBYTECODE=1 pytest -q tests/integration/defs/accuracy/test_llm_api_autodeploy.py::TestQwen3_5_35B_MoE::test_bf16 -s

Result from the full-model MMMU run:

• mmmu_val average accuracy: 40.11
• test status: 1 passed

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)

• 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.

Summary by CodeRabbit

Release Notes

• New Features

  • Enhanced Qwen 3.5 MoE model support with multimodal position handling capabilities.
  • Added automatic model factory registration for improved deployment configuration.

• Tests

  • Comprehensive multimodal model test coverage including position computation and caching scenarios.
  • New validation tests for deployment transform initialization and sharding configuration.

[bmarimuthu-nv]

Implementation update for the latest Qwen3.5 MoE AutoDeploy change set:

Plan executed:

1. Keep request-level multimodal span construction Qwen-specific by deriving the exact multimodal spans from tokenized input_ids in the Qwen model path.
2. Make the VLM wrapper chunk-aware so multimodal 3D position construction follows the real chunked AutoDeploy execution layout.
3. Move mRoPE delta handling to a proper per-request cache by adding a tiny per-slot mrope_delta_cache state resource and using cached read/write custom ops in the wrapper.
4. Preserve warmup/export stability by treating metadata-free prefill warmups as a zero-delta no-op.
5. Fix the remaining init-time failure for flattened cached position_ids by expanding request-level deltas to per-token deltas with cu_seqlen in the text-only/decode fallback.
6. Re-enable the full-depth Qwen3.5 35B MoE accuracy path after e2e revalidation.

Validation summary:

• Targeted unit coverage passed for mixed image+video request-level delta matching, cache prefill/decode behavior, and flattened cached prefill delta expansion.
• Full-model AutoDeploy MMMU accuracy run completed successfully after the fallback fix.
• Final full-model result: mmmu_val average accuracy 40.11.
• Full-model pytest result: 1 passed.

Relevant local commands used for validation:

• pytest -q tests/unittest/_torch/auto_deploy/unit/singlegpu/models/test_qwen3_5_moe.py -k "mrope_delta_matches_compute_mrope_positions_for_mixed_items or mrope_delta_with_cache_writes_prefill_and_reads_decode or flattened_prefill_with_request_deltas" -p no:cacheprovider
• TRTLLM_ACCURACY_NO_REFERENCE=1 TMPDIR=/tmp/trtqwen HF_HOME=/tmp/trtqwen/hf_home PYTHONDONTWRITEBYTECODE=1 pytest -q tests/integration/defs/accuracy/test_llm_api_autodeploy.py::TestQwen3_5_35B_MoE::test_bf16 -s

Latest branch commit: 7bf955df4e ([None][fix] cache Qwen3.5 mRoPE deltas in AutoDeploy).

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38845 [ run ] triggered by Bot. Commit: de6312f Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38846 [ run ] triggered by Bot. Commit: de6312f Link to invocation

[tensorrt-cicd]

PR_Github #38846 [ run ] completed with state SUCCESS. Commit: de6312f
/LLM/main/L0_MergeRequest_PR pipeline #30156 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

Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38888 [ run ] triggered by Bot. Commit: aea2358 Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38888 [ run ] completed with state SUCCESS. Commit: aea2358
/LLM/main/L0_MergeRequest_PR pipeline #30196 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

Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38907 [ run ] triggered by Bot. Commit: 318cede Link to invocation

[tensorrt-cicd]

PR_Github #38907 [ run ] completed with state FAILURE. Commit: 318cede
/LLM/main/L0_MergeRequest_PR pipeline #30214 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

Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38925 [ run ] triggered by Bot. Commit: c48de0b Link to invocation

[tensorrt-cicd]

PR_Github #38925 [ run ] completed with state DISABLED
CI server is currently disabled for scheduled maintenance. Estimated completion time: 9 PM PST on 3/14.

Link to invocation

[bmarimuthu-nv]

/bot run --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1" --disable-fail-fast

[tensorrt-cicd]

PR_Github #38961 [ run ] triggered by Bot. Commit: c48de0b Link to invocation

[tensorrt-cicd]

PR_Github #38961 [ run ] completed with state SUCCESS. Commit: c48de0b
/LLM/main/L0_MergeRequest_PR pipeline #30243 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

Link to invocation

[suyoggupta]

/bot run

[tensorrt-cicd]

PR_Github #40039 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40039 [ run ] completed with state SUCCESS. Commit: d4c4de9
/LLM/main/L0_MergeRequest_PR pipeline #31194 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

Link to invocation

[bmarimuthu-nv]

/bot run --disable-fail-fast --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

[tensorrt-cicd]

PR_Github #40098 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40098 [ run ] completed with state SUCCESS. Commit: d4c4de9
/LLM/main/L0_MergeRequest_PR pipeline #31249 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

Link to invocation

[bmarimuthu-nv]

/bot run --disable-fail-fast --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

[tensorrt-cicd]

PR_Github #40145 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40145 [ run ] completed with state SUCCESS. Commit: d4c4de9
/LLM/main/L0_MergeRequest_PR pipeline #31291 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

Link to invocation

[bmarimuthu-nv]

/bot run --disable-fail-fast --reuse-test --extra-stage "DGX_B200-4_GPUs-AutoDeploy-1, DGX_H100-4_GPUs-AutoDeploy-1"

[tensorrt-cicd]

PR_Github #40180 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40180 [ run ] completed with state SUCCESS. Commit: d4c4de9
/LLM/main/L0_MergeRequest_PR pipeline #31323 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

Link to invocation

[bmarimuthu-nv]

/bot run --reuse-test --stage-list "DGX_H100-4_GPUs-AutoDeploy-1"

[tensorrt-cicd]

PR_Github #40332 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40331 [ run ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40332 [ run ] completed with state ABORTED. Commit: d4c4de9

Link to invocation

[tensorrt-cicd]

PR_Github #40331 [ run ] completed with state SUCCESS. Commit: d4c4de9
/LLM/main/L0_MergeRequest_PR pipeline #31439 (Partly Tested) completed with status: 'SUCCESS'

CI Report

Link to invocation

[bmarimuthu-nv]

/bot help

[github-actions]

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

Details

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental) --high-priority]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id (OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test (OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--test-backend "pytorch, cpp" (OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log (OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug (OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

--high-priority (OPTIONAL) : Run the pipeline with high priority. This option is restricted to authorized users only and will route the job to a high-priority queue.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[bmarimuthu-nv]

/bot skip --m "all tests passing"

[github-actions]

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

Details

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental) --high-priority]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id (OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test (OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--test-backend "pytorch, cpp" (OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log (OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug (OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

--high-priority (OPTIONAL) : Run the pipeline with high priority. This option is restricted to authorized users only and will route the job to a high-priority queue.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[bmarimuthu-nv]

/bot skip --comment "all tests passing"

[tensorrt-cicd]

PR_Github #40360 [ skip ] triggered by Bot. Commit: d4c4de9 Link to invocation

[tensorrt-cicd]

PR_Github #40360 [ skip ] completed with state SUCCESS. Commit: d4c4de9
Skipping testing for commit d4c4de9

Link to invocation