[https://nvbugs/5744427][fix] Fix accuracy test OOM

[brb-nv]

Description

Issue:
Gemma3 27B VLM goes OOM on H100 PCIe with 80GB of memory.

Root cause:

• Original test runs on a tight memory budget with BF16 weights taking up 52GB memory.

[12/19/2025-22:52:36] [TRT-LLM] [I] Use 51.61 GB for model weights.
[12/19/2025-23:03:52] [TRT-LLM] [I] Memory dynamically allocated during inference (inside torch) in memory usage profiling: 5.16 GiB
[12/19/2025-23:03:52] [TRT-LLM] [I] Memory used outside torch (e.g., NCCL and CUDA graphs) in memory usage profiling: 16.55 GiB
[12/19/2025-23:03:52] [TRT-LLM] [I] Peak memory during memory usage profiling (torch + non-torch): 74.41 GiB, available KV cache memory when calculating max tokens: 10.15 GiB, fraction is set 0.6, kv size is 507904. device total memory 79.18 GiB, , tmp kv_mem 12.15 GiB
[12/19/2025-23:03:52] [TRT-LLM] [I] Estimated max memory in KV cache : 10.15 GiB
[12/19/2025-23:03:52] [TRT-LLM] [W] Both free_gpu_memory_fraction and max_tokens are set (to 0.6 and 21464 with free memory 22.04754638671875GiB of total memory 79.1788330078125GiB, respectively). The smaller value will be used.
[TensorRT-LLM][INFO] Max KV cache blocks per sequence: 273 [window size=8705], tokens per block=32, primary blocks=670, secondary blocks=0, max sequence length=8705
[TensorRT-LLM][INFO] Number of tokens per block: 32.
[TensorRT-LLM][INFO] [MemUsageChange] Allocated 10.14 GiB for max tokens in paged KV cache (21440).
[12/19/2025-23:03:53] [TRT-LLM] [I] max_seq_len=8705, max_num_requests=16, max_num_tokens=25600, max_batch_size=16

• Main issue is vision encoder's activation don't seem to be taken into account in initial memory profiling.

Fix:

• I think the principled fix would be to make a change in tensorrt_llm/_torch/models/modeling_gemma3vl.py from this commit:
  2325867
  This results in creation of dummy multimodal requests and subsequent memory estimation of vision encoder activations through this function:

  TensorRT-LLM/tensorrt_llm/_torch/pyexecutor/_util.py

  Line 199 in 2325867

  def _create_dummy_context_requests(

• But this is causing a large spike in test time because the multimodal request creation takes quite some time.
• I'm afraid this will result in timeout issues in CI.
• So, I decided to turn the knobs of max_num_tokens and free_gpu_memory_fraction in the test for now. Verified that the test now passes on H100 PCIe.

Test Coverage

$ pytest tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py::TestGemma3_27BInstruct::test_auto_dtype -s -v

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

/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)]

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.

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md
and the scripts/test_to_stage_mapping.py helper.

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.

Summary by CodeRabbit

• Tests
  • Updated test configuration parameters to optimize resource allocation and computational efficiency.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR adjusts test configuration parameters for the Gemma 3 27B Instruct multimodal LLM test. The maximum token budget is reduced from 25600 to 12800, and the free GPU memory fraction is decreased from 0.6 to 0.4.

Changes

Cohort / File(s)	Change Summary
Test Configuration Updates tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py	Reduced MAX_NUM_TOKENS from 25600 to 12800; lowered free_gpu_memory_fraction in KvCacheConfig from 0.6 to 0.4 for Gemma 3 27B Instruct test

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Pre-merge checks and finishing touches

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title correctly identifies the fix for an OOM issue in an accuracy test, which matches the main change in the changeset.
Description check	✅ Passed	PR description comprehensively explains the issue, root cause, attempted fixes, and rationale for the chosen solution with supporting evidence and test verification.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (3)

tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py (3)

222-222: Verify that reduced token budget maintains adequate test coverage.

The 50% reduction in MAX_NUM_TOKENS (from 25600 to 12800) is substantial and may limit test coverage for longer multimodal sequences. While this interim fix addresses the OOM issue, please confirm that the MMMU evaluation still exercises meaningful scenarios with this reduced budget.

Consider adding a comment referencing the tracking bug (https://nvbugs/5744427) and noting that this is a temporary workaround until vision encoder activations are properly accounted for in memory estimation.

Suggested documentation improvement

+    # NOTE: Reduced from 25600 to avoid OOM on H100 PCIe (nvbugs/5744427).
+    # This is an interim fix until vision encoder activations are included in memory estimation.
     MAX_NUM_TOKENS = 12800

---

229-233: Conservative memory fraction—verify this doesn't over-constrain the test.

The free_gpu_memory_fraction of 0.4 is the lowest among all multimodal tests in this file (others range from 0.6–0.8). Combined with the 50% reduction in MAX_NUM_TOKENS, this represents a significant constraint on available memory.

While this ensures the test passes on H100 PCIe 80GB, please verify:

1. Whether both reductions are necessary, or if adjusting only one parameter would suffice
2. That KV cache capacity remains adequate for the test workload

Consider documenting why Gemma3 27B requires more conservative settings than other large VLM models.

---

219-247: Pragmatic interim fix—consider tracking long-term solution.

The memory constraint adjustments are a reasonable workaround for the OOM issue on H100 PCIe. The PR description clearly explains the root cause (vision encoder activations not accounted for in memory estimation) and references the principled fix (commit 2325867).

To aid future maintenance, consider:

1. Adding inline comments referencing nvbugs/5744427 to explain why Gemma3 has uniquely conservative settings
2. Adding a TODO or tracking issue comment for the principled fix to prevent this workaround from becoming permanent

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 3f25db9 and 251d665.

📒 Files selected for processing (1)

• tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py (2 hunks)

🧰 Additional context used

📓 Path-based instructions (2)

**/*.py

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.py: Code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces. Do not use tabs
Always maintain the namespace when importing in Python, even if only one class or function from a module is used
Python files should use snake_case naming: some_file.py
Python classes should use PascalCase naming: class SomeClass
Python functions and methods should use snake_case naming: def my_awesome_function():
Python local variables should use snake_case naming: my_variable = ...
Python variable names that start with a number should be prefixed with 'k': k_99th_percentile = ...
Python global variables should use upper snake_case with prefix 'G': G_MY_GLOBAL = ...
Python constants should use upper snake_case naming: MY_CONSTANT = ...
Avoid shadowing variables declared in an outer scope in Python
Initialize all externally visible members of a Python class in the constructor
For Python interfaces that may be used outside a file, prefer docstrings over comments
Python comments should be reserved for code within a function, or interfaces that are local to a file
Use Google style docstrings in Python for classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except to the smallest set of errors possible
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible, using the else block for logic

Files:

• tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

**/*.{cpp,h,cu,cuh,py}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the year of its latest meaningful modification

Files:

• tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

🧠 Learnings (2)

📓 Common learnings

Learnt from: venkywonka
Repo: NVIDIA/TensorRT-LLM PR: 6029
File: .github/pull_request_template.md:45-53
Timestamp: 2025-08-27T17:50:13.264Z
Learning: For PR templates in TensorRT-LLM, avoid suggesting changes that would increase developer overhead, such as converting plain bullets to mandatory checkboxes. The team prefers guidance-style bullets that don't require explicit interaction to reduce friction in the PR creation process.

📚 Learning: 2025-08-14T21:04:50.248Z

Learnt from: thorjohnsen
Repo: NVIDIA/TensorRT-LLM PR: 6910
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-14T21:04:50.248Z
Learning: In KV cache onboarding logic during prefill in cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, when calculating which blocks fall within the attention window, use getTokensPerBlock() to advance token indices rather than block->getUniqueTokens().size(), because the calculation needs to consider the post-prefill state where blocks will be filled to capacity, not their current token count.

Applied to files:

• tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: Pre-commit Check

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #29211 [ run ] triggered by Bot. Commit: a821058

[tensorrt-cicd]

PR_Github #29211 [ run ] completed with state SUCCESS. Commit: a821058
/LLM/main/L0_MergeRequest_PR pipeline #22414 completed with status: 'FAILURE'

⚠️ 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

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #29253 [ run ] triggered by Bot. Commit: a821058

[tensorrt-cicd]

PR_Github #29253 [ run ] completed with state SUCCESS. Commit: a821058
/LLM/main/L0_MergeRequest_PR pipeline #22456 completed with status: 'FAILURE'

⚠️ 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

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #29275 [ run ] triggered by Bot. Commit: cd61925

[brb-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #29277 [ run ] triggered by Bot. Commit: 3fb2c50

[tensorrt-cicd]

PR_Github #29277 [ run ] completed with state SUCCESS. Commit: 3fb2c50
/LLM/main/L0_MergeRequest_PR pipeline #22476 completed with status: 'SUCCESS'