[None][feat] Graceful Error Handling for Guided Decoder

[jellysnack]

Summary by CodeRabbit

• Bug Fixes
  • Improved guided decoding error handling to gracefully manage per-request failures without stopping batch processing.
  • Enhanced error reporting now includes request-specific failure details and messages.
  • Strengthened error propagation through the execution pipeline for better visibility into decoding issues.
  • Increased robustness when individual requests encounter errors during guided decoding operations.

Description

This PR adds graceful error handling for the guided decoder component, preventing executor loop crashes when grammar matching errors occur (e.g., invalid JSON schema). Failed requests are now terminated individually with detailed error messages, allowing the executor to continue serving other requests.

Failed requests now receive detailed error messages:

Guided decoding error: <error details>

This allows to:

• Identify guided decoding failures specifically
• Wrap them in appropriate HTTP responses (e.g., 400 Invalid Argument)
• Provide actionable feedback to users

Speculative decoding error handling is incomplete. Exceptions during guided decoding are caught, but failed requests continue processing without guided decoding applied rather than being terminated with an error response. The complexity of drafting in different modes (one-model / two-model paths) makes it non-obvious how to properly handle guided decoder errors in the speculative pipeline without introducing additional issues.

If needed, I'm happy to extend this PR to cover speculative decoding with guidance from maintainers on the preferred approach. Alternatively, this could be addressed in a separate PR by the team if that's more appropriate.

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.

[coderabbitai]

📝 Walkthrough

Walkthrough

Modified guided decoder to collect and return failed requests as List[Tuple[int, str]] instead of raising exceptions. Updated executor to capture these failures and process them via a new error handling helper. Changes enable per-request failure tracking rather than batch-wide exception termination.

Changes

Cohort / File(s)	Summary
Guided Decoder Error Collection tensorrt_llm/_torch/pyexecutor/guided_decoder.py	Updated _build, build, execute, execute_draft_batch and hostfunc variants to return List[Tuple[int, str]] of failed requests. Introduced per-request try/except blocks to collect failures instead of raising. Added None-checks for matcher during guided decoding. Replaced direct exception raises with failure recording and logging. Changes applied to both GuidedDecoder and CapturableGuidedDecoder classes.
Executor Integration tensorrt_llm/_torch/pyexecutor/py_executor.py	Added new _handle_guided_decoder_errors helper to map per-request guided-decoder failures to standard error handling. Updated _handle_errors to accept Union[str, List[str]] for per-request error messages with length-mismatch guards. Integrated guided-decoder failure handling into all executor loop variants (pp, standard, overlap) by capturing failures from GuidedDecoder.execute() and processing post-sampling.

Sequence Diagram(s)

sequenceDiagram
    participant Executor
    participant GuidedDecoder
    participant ErrorHandler
    
    rect rgb(240, 248, 255)
    Note over Executor,ErrorHandler: New: Per-Request Error Handling
    end
    
    Executor->>GuidedDecoder: execute(logits, d2t)
    activate GuidedDecoder
    Note over GuidedDecoder: Per-request try/except<br/>Collect failures instead<br/>of raising exceptions
    GuidedDecoder->>GuidedDecoder: Process each request<br/>Log errors, record failures
    GuidedDecoder-->>Executor: List[Tuple[int, str]]<br/>(failed_requests)
    deactivate GuidedDecoder
    
    Executor->>Executor: Apply bitmasks (sampling)
    
    rect rgb(255, 240, 240)
    Note over Executor,ErrorHandler: Error Processing Phase
    end
    
    Executor->>ErrorHandler: _handle_guided_decoder_errors<br/>(scheduled_batch, failed_requests)
    activate ErrorHandler
    ErrorHandler->>ErrorHandler: Map per-request failures<br/>to error responses
    ErrorHandler->>ErrorHandler: Call _handle_errors<br/>with per-request messages
    deactivate ErrorHandler
    
    ErrorHandler-->>Executor: Error responses created

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

• guided_decoder.py: Multiple coordinated method signature changes across two classes with modified control flow logic. Requires verification that per-request error collection doesn't lose context and that all code paths (draft/non-draft) consistently return failure lists.
• py_executor.py: New error handling helper and updated _handle_errors signature with Union type handling. Cross-file integration complexity: verify that captured failures are correctly mapped to requests and that the order/indexing aligns between scheduled batch and failure tuples.
• Key areas requiring attention:
  • Verify per-request try/except blocks in _build correctly capture all exception types and maintain request-to-error-message mapping
  • Confirm failure list propagation through all executor loop variants maintains consistency
  • Check length-mismatch guards in _handle_errors work correctly for varying batch sizes and failure counts
  • Validate that None-matcher checks don't inadvertently skip legitimate error handling

Pre-merge checks and finishing touches

❌ Failed checks (1 warning, 1 inconclusive)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 15.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.
Description check	❓ Inconclusive	PR description covers the main objective and reasoning but lacks test coverage details and leaves the checklist items unchecked despite claiming completion.	Specify what tests validate the guided decoder error handling changes. Clearly mark or complete the PR checklist items to confirm coding guidelines, test cases, and documentation have been properly addressed.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title check	✅ Passed	The PR title '[None][feat] Graceful Error Handling for Guided Decoder' clearly and specifically describes the main change: adding graceful error handling to the guided decoder component.

✨ 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.}

[juney-nvidia]

Hi @syuoni @sunnyqgg can you help review this PR?

Thanks
June

[sunnyqgg]

LGTM

[jellysnack]

@syuoni Hi! Could you please take a look at this PR when you have a moment?