[Perf] Optimize flashinfer sampling via 'filter_apply_order="joint"'

[HirokenOvo]

Summary by CodeRabbit

• Bug Fixes
  • Improved sampling behavior by adjusting how top-k and top-p filtering are applied during token generation, potentially enhancing sampling quality.

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

When min_tokens is set in Sampling Params, using FlashInfer's top_k_top_p_sampling_from_logits kernel causes a significant increase in TPOT and TTFT. This is because the default filter_apply_order="top_k_first" is sensitive to -inf logit values, causing severe performance degradation when they are present. Switching the order to "joint" fixes this regression and additionally improves performance in general scenarios where min_tokens is not set.

For some torch profiler traces and Accuracy test of what this looks like, see an identical bug in vLLM: vllm-project/vllm#29979

this PR. With min_tokens (introduces -inf):

# server 
trtllm-serve serve /workspace/models/Qwen3-32B/
# client
aiperf profile --model /workspace/models/Qwen3-32B/ --endpoint-type chat --url localhost:8000 --streaming --concurrency 1 --request-count 256 --warmup-request-count 0 --isl 50 --osl 200  --extra-inputs min_tokens:200  --extra-inputs top_k:20 --extra-inputs top_p:0.8   --artifact-dir /tmp --profile-export-file qwen3-32b

                                           NVIDIA AIPerf | LLM Metrics                                           
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━┓
┃                               Metric ┃      avg ┃      min ┃      max ┃      p99 ┃      p90 ┃      p50 ┃  std ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━┩
│             Time to First Token (ms) │    40.10 │    38.35 │    44.13 │    43.82 │    41.05 │    39.84 │ 1.50 │
│            Time to Second Token (ms) │    24.36 │    23.97 │    24.61 │    24.60 │    24.49 │    24.37 │ 0.16 │
│                 Request Latency (ms) │ 4,943.07 │ 4,941.66 │ 4,946.19 │ 4,946.00 │ 4,944.31 │ 4,942.87 │ 1.26 │
│             Inter Token Latency (ms) │    24.64 │    24.63 │    24.64 │    24.64 │    24.64 │    24.64 │ 0.00 │
│     Output Token Throughput Per User │    40.59 │    40.58 │    40.60 │    40.59 │    40.59 │    40.59 │ 0.00 │
│                    (tokens/sec/user) │          │          │          │          │          │          │      │
│      Output Sequence Length (tokens) │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │ 0.00 │
│       Input Sequence Length (tokens) │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │ 0.00 │
│ Output Token Throughput (tokens/sec) │    40.43 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│    Request Throughput (requests/sec) │     0.20 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│             Request Count (requests) │    10.00 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
└──────────────────────────────────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────┘

main. With min_tokens (introduces -inf):

                                           NVIDIA AIPerf | LLM Metrics                                           
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━┓
┃                               Metric ┃      avg ┃      min ┃      max ┃      p99 ┃      p90 ┃      p50 ┃  std ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━┩
│             Time to First Token (ms) │    81.87 │    77.23 │    95.02 │    93.93 │    84.04 │    80.53 │ 4.69 │
│            Time to Second Token (ms) │    39.57 │    39.18 │    40.34 │    40.32 │    40.16 │    39.34 │ 0.41 │
│                 Request Latency (ms) │ 7,125.91 │ 7,119.39 │ 7,143.17 │ 7,142.23 │ 7,133.76 │ 7,123.78 │ 7.17 │
│             Inter Token Latency (ms) │    35.40 │    35.37 │    35.43 │    35.43 │    35.42 │    35.39 │ 0.02 │
│     Output Token Throughput Per User │    28.25 │    28.22 │    28.27 │    28.27 │    28.27 │    28.25 │ 0.02 │
│                    (tokens/sec/user) │          │          │          │          │          │          │      │
│      Output Sequence Length (tokens) │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │ 0.00 │
│       Input Sequence Length (tokens) │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │ 0.00 │
│ Output Token Throughput (tokens/sec) │    28.05 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│    Request Throughput (requests/sec) │     0.14 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│             Request Count (requests) │    10.00 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
└──────────────────────────────────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────┘

main. Without min_tokens (no -inf):

aiperf profile --model /workspace/models/Qwen3-32B/ --endpoint-type chat --url localhost:8000 --streaming --concurrency 1 --request-count 256 --warmup-request-count 0 --isl 50 --osl 200   --extra-inputs top_k:20 --extra-inputs top_p:0.8   --artifact-dir /tmp --profile-export-file qwen3-32b

                                           NVIDIA AIPerf | LLM Metrics                                           
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━┓
┃                               Metric ┃      avg ┃      min ┃      max ┃      p99 ┃      p90 ┃      p50 ┃  std ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━┩
│             Time to First Token (ms) │    40.84 │    38.58 │    47.18 │    46.69 │    42.24 │    40.35 │ 2.33 │
│            Time to Second Token (ms) │    25.62 │    25.16 │    25.87 │    25.87 │    25.83 │    25.67 │ 0.21 │
│                 Request Latency (ms) │ 5,177.12 │ 5,171.70 │ 5,185.02 │ 5,184.60 │ 5,180.82 │ 5,176.47 │ 3.91 │
│             Inter Token Latency (ms) │    25.81 │    25.79 │    25.83 │    25.83 │    25.83 │    25.81 │ 0.01 │
│     Output Token Throughput Per User │    38.74 │    38.72 │    38.77 │    38.77 │    38.77 │    38.74 │ 0.02 │
│                    (tokens/sec/user) │          │          │          │          │          │          │      │
│      Output Sequence Length (tokens) │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │   200.00 │ 0.00 │
│       Input Sequence Length (tokens) │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │    50.00 │ 0.00 │
│ Output Token Throughput (tokens/sec) │    38.60 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│    Request Throughput (requests/sec) │     0.19 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
│             Request Count (requests) │    10.00 │      N/A │      N/A │      N/A │      N/A │      N/A │  N/A │
└──────────────────────────────────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────────┴──────┘

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)

• 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.

[coderabbitai]

📝 Walkthrough

Walkthrough

A parameter configuration change in the flashinfer sampling utility: the filter_apply_order argument for TopKTopPSampleOnly was updated from "top_k_first" to "joint" to alter how top-k and top-p filtering are applied during joint processing.

Changes

Cohort / File(s)	Summary
Flashinfer sampling configuration tensorrt_llm/_torch/pyexecutor/sampling_utils_flashinfer.py	Updated filter_apply_order parameter from "top_k_first" to "joint" in flashinfer.sampling.top_k_top_p_sampling_from_logits call for TopKTopPSampleOnly

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

• Single parameter change in one file
• No logic or control flow modifications
• Clear impact: changes filtering application order during sampling

Pre-merge checks and finishing touches

❌ Failed checks (2 warnings)

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.
Description check	⚠️ Warning	The PR description is incomplete. While the author provided detailed context about the performance issue and benchmark results, critical template sections like PR title and formal description structure are missing.	Add a proper PR title following the template format [ticket][type] Summary, fill in the Description section explaining the issue and solution, and clearly list relevant tests in the Test Coverage section.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly indicates a performance optimization by changing the 'filter_apply_order' parameter to 'joint' in flashinfer sampling, directly matching the code change.

✨ 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.}