[TRTLLM-12624][ci] Drop tensorrt_llm/llmapi/ from multi-GPU trigger list

[QiJune]

Summary

Narrow the multi-GPU trigger rule in getMultiGpuFileChanged by replacing the blanket tensorrt_llm/llmapi/ prefix with 5 explicit files that correspond to multi-rank / multi-node / disagg orchestration entry points.

The directory prefix triggered MGPU pipelines on PRs that only modified public-API surface (llm.py, llm_args.py), parsers (reasoning_parser.py, tokenizer.py), telemetry, or single-field config defaults — none of which exercise multi-rank code paths.

Measured impact (12-week window on main, 995 commits)

	Today	After
MGPU pipeline triggers	365	343
Hit rate	36.7 %	34.5 %
llmapi-solo MGPU triggers (true positives)	22 (0)	0

• 22 fewer MGPU pipeline runs per 12 weeks (~95 / year), all on the scarcest multi-GPU runner pool.
• 0 measured missed regressions in the window. The closest candidate (bd14845e094174de6bf52f4f5509264ef975d12a) was a None → 60 s default-value tightening on kv_transfer_timeout_ms; L0 disagg tests transfer in ms and never reach 60 s.

Files kept as explicit MGPU triggers

File	Role
tensorrt_llm/llmapi/mpi_session.py	MPI session lifecycle, multi-rank spawn
tensorrt_llm/llmapi/mgmn_leader_node.py	Multi-GPU multi-node leader role
tensorrt_llm/llmapi/mgmn_worker_node.py	Multi-GPU multi-node worker role
tensorrt_llm/llmapi/disagg_utils.py	Disagg serving utilities
tensorrt_llm/llmapi/trtllm-llmapi-launch	Multi-rank launch script

These 5 are touched by only 3 commits in the 12-week window, and all 3 also trigger MGPU via executor/, pyexecutor/, or serve/. The carve-out therefore adds 0 marginal triggers in the measured window, but is retained as forward-looking insurance for the rare PR that only touches multi-rank orchestration entry points.

@coderabbitai summary

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

To see a list of available CI bot commands, please comment /bot help.

[QiJune]

/bot skip --comment "trivial changes"

[tensorrt-cicd]

PR_Github #47807 [ skip ] triggered by Bot. Commit: e67e4c3 Link to invocation

[tensorrt-cicd]

PR_Github #47807 [ skip ] completed with state FAILURE. Commit: e67e4c3

Link to invocation

[QiJune]

/bot skip --comment "trivial changes"

[tensorrt-cicd]

PR_Github #47818 [ skip ] triggered by Bot. Commit: e67e4c3 Link to invocation

[tensorrt-cicd]

PR_Github #47818 [ skip ] completed with state SUCCESS. Commit: e67e4c3
Skipping testing for commit e67e4c3

Link to invocation