[#11037][fix] Fix proto-to-SamplingParams conversion bugs and add gRPC tests

[CatherineSue]

Summary by CodeRabbit

• Bug Fixes

  • Fixed beam search parameter handling to correctly enable and configure beam search functionality.
  • Corrected seed parameter assignment for proper randomization control.
  • Improved stop words and bad words processing for more reliable decoding.

• Tests

  • Expanded test coverage for sampling parameter conversion and gRPC service functionality.
  • Added comprehensive end-to-end tests for generate flows, health checks, and model information retrieval.

Description

Fix 5 bugs in create_sampling_params_from_proto() where kwargs didn't match SamplingParams dataclass fields (slots=True, kw_only=True), causing TypeError at runtime:

• beam_width → use_beam_search=True + best_of=beam_width
• random_seed → seed
• stop_words/bad_words as kwargs → set _stop_word_ids/_bad_word_ids post-construction
• guided_decoding_params → guided_decoding
• GuidedDecodingParams(json_schema=...) → GuidedDecodingParams(json=...)

Add comprehensive proto conversion tests and e2e gRPC service tests as requested in #11037.

Test Coverage

• TestComprehensiveSamplingParamsConversion (4 tests, no GPU):
  • test_all_sampling_config_fields — sets every proto field, verifies every SamplingParams field
  • test_end_id_minus_one_sets_ignore_eos
  • test_defaults_when_fields_unset
  • test_guided_decoding_all_types — JSON, JSON Schema, Regex, EBNF Grammar
• TestGrpcServiceEndToEnd (6 tests, GPU required, TinyLlama-1.1B):
  • test_generate_non_streaming — verifies complete response with token IDs
  • test_generate_streaming — verifies delta chunks reassemble to match complete output
  • test_health_check
  • test_abort_nonexistent_request
  • test_get_model_info
  • test_get_server_info

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.

[coderabbitai]

📝 Walkthrough

Walkthrough

The changes refactor SamplingParams creation in grpc_request_manager.py to use pre-tokenized word IDs for stop/bad words, update beam search handling via use_beam_search and best_of parameters, and rename guided decoding configurations. Tests are expanded with comprehensive SamplingParams conversion coverage and end-to-end gRPC service tests including fixtures and async helpers.

Changes

Cohort / File(s)	Summary
SamplingParams Mapping tensorrt_llm/grpc/grpc_request_manager.py	Refactored SamplingParams construction: beam_width now maps to use_beam_search + best_of; seed parameter renamed; stop/bad words converted to pre-tokenized word IDs injected after object construction; guided_decoding_params renamed to guided_decoding with schema key adjustments.
Test Coverage Expansion tests/unittest/llmapi/test_grpc.py	Updated SamplingParams assertions to reflect new beam search and guided decoding mappings. Added TestComprehensiveSamplingParamsConversion class covering all SamplingConfig fields, embedding bias conversion, and end_id handling. Introduced TestGrpcServiceEndToEnd with fixtures (grpc_service), async helpers, and end-to-end tests for generate flows, health checks, and model info retrieval.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 69.23% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main changes: fixing bugs in proto-to-SamplingParams conversion and adding comprehensive test coverage.
Description check	✅ Passed	The description comprehensively covers the five specific bugs fixed, test coverage details with test counts and scopes, and confirms PR checklist completion.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ 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]

@QiJune Hi Jun, can you help review this PR?

Thanks
June

[QiJune]

/bot run

[QiJune]

LGTM

[tensorrt-cicd]

PR_Github #34863 [ run ] triggered by Bot. Commit: 2fd36b3

[tensorrt-cicd]

PR_Github #34863 [ run ] completed with state SUCCESS. Commit: 2fd36b3
/LLM/main/L0_MergeRequest_PR pipeline #26897 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

[QiJune]

/bot run

[tensorrt-cicd]

PR_Github #34932 [ run ] triggered by Bot. Commit: 2fd36b3

[tensorrt-cicd]

PR_Github #34932 [ run ] completed with state SUCCESS. Commit: 2fd36b3
/LLM/main/L0_MergeRequest_PR pipeline #26949 completed with status: 'SUCCESS'