[TRTLLM-7963][fix] Several improvements of autotuning quality

[hyukn]

This PR skips the shape profile generating process if the profile has already been found in the cache under tuning mode.

This is a prerequisite for nested autotuning because host overhead might be included during the profiling process of the high-level op.

Also, this PR:

• Enable the profiling with CUDA graph as the default profiling method.
• Apply a heuristic method to cut off repeat times of profiling according to the measured time of a few runs. The number of tuning processes will become:
  • if execution is long: 1 (pure warmup) + 2 (fewer repeats) * 2 (graph replay if using CUDA graph). Directly use the time measured in the "fewer repeats" stage.
  • if execution is short: 1 (pure warmup) + 2 (fewer repeats) * 2 + 10 (common repeats) * 2

Summary by CodeRabbit

• Refactor
  • Autotuner now performs an upfront cache lookup and uses cached results to avoid redundant profiling.
  • Tuning mode short-circuits on cache hits to skip extra work; non-tuning mode reuses prefetched cache results and logs cache misses.
  • Overall performance and efficiency of autotuning improved.

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

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.

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.

[hyukn]

/bot run

[coderabbitai]

📝 Walkthrough

Walkthrough

Early cache lookup in choose_one now retrieves is_cache_hit, best_runner_id, best_tactic, and min_time before branching. Tuning mode short-circuits and returns cached results on hit. Non-tuning code uses the pre-fetched cache state and logs cache misses without repeating the lookup.

Changes

Cohort / File(s)	Change Summary
Cache lookup optimization in autotuner tensorrt_llm/_torch/autotuner.py	Perform cache search at function start to obtain is_cache_hit, best_runner_id, best_tactic, and min_time. On tuning-mode cache hit, return cached (runner, tactic) immediately. Non-tuning path uses pre-fetched cache result and logs misses; removed redundant cache lookup. Minor validation of runners/tactics retained.

Sequence Diagram

sequenceDiagram
    actor Caller
    participant choose_one as choose_one()
    participant Cache as profiling_cache

    Caller->>choose_one: invoke(inputs, config)
    activate choose_one

    rect rgb(230,240,255)
      Note over choose_one,Cache: Early cache lookup (new)
      choose_one->>Cache: lookup(key)
      Cache-->>choose_one: is_cache_hit, best_runner_id, best_tactic, min_time
    end

    alt is_cache_hit AND tuning_mode
      rect rgb(200,255,200)
        Note over choose_one: Short-circuit return (new)
        choose_one-->>Caller: return cached (runner, tactic)
      end
    else
      alt tuning_mode
        rect rgb(255,245,200)
          Note over choose_one: Perform profiling
          choose_one->>choose_one: profile candidates
          choose_one-->>Caller: return profiled (runner, tactic)
        end
      else not tuning_mode
        rect rgb(255,230,230)
          Note over choose_one: Use pre-fetched cache state
          choose_one->>choose_one: use cached values or fallback
          choose_one-->>Caller: return selected (runner, tactic)
        end
      end
    end

    deactivate choose_one

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

• Verify early cache lookup populates all variables used later.
• Confirm tuning-mode early return condition is correct and doesn't skip necessary validation.
• Ensure non-tuning paths correctly use pre-fetched cache state and log misses only once.
• Check retained validation of runner/tactic presence and correct fallback behavior.

Pre-merge checks and finishing touches

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	PR description lacks critical sections: 'Description' and 'Test Coverage' sections are empty templates, and PR Checklist items are incomplete.	Fill in the 'Description' section with clear explanation of the issue and solution, and complete 'Test Coverage' section with relevant tests that validate the changes.
Title check	❓ Inconclusive	The PR title mentions 'Several improvements of autotuning quality' but the actual change focuses on avoiding redundant profile generation by skipping duplicate shape profile generation when a profile is cached during tuning mode.	The title is vague about the specific improvement. Consider a more precise title like '[TRTLLM-7963][fix] Skip redundant shape profile generation on cache hit during tuning' to clearly describe the main change.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

✨ 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.}

[tensorrt-cicd]

PR_Github #25282 [ run ] triggered by Bot. Commit: 5a1f1b2

[tensorrt-cicd]

PR_Github #25282 [ run ] completed with state FAILURE. Commit: 5a1f1b2
/LLM/main/L0_MergeRequest_PR pipeline #19127 completed with status: 'FAILURE'

[hyukn]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #25290 [ run ] triggered by Bot. Commit: 0306cc1

[tensorrt-cicd]

PR_Github #25290 [ run ] completed with state FAILURE. Commit: 0306cc1
/LLM/main/L0_MergeRequest_PR pipeline #19133 completed with status: 'FAILURE'

[hyukn]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #25302 [ run ] triggered by Bot. Commit: c0b4079

[tensorrt-cicd]

PR_Github #25302 [ run ] completed with state FAILURE. Commit: c0b4079
/LLM/main/L0_MergeRequest_PR pipeline #19141 completed with status: 'FAILURE'

[hyukn]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #25333 [ run ] triggered by Bot. Commit: d2de147

[tensorrt-cicd]

PR_Github #25333 [ run ] completed with state SUCCESS. Commit: d2de147
/LLM/main/L0_MergeRequest_PR pipeline #19163 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.