Skip to content

ci: decouple stage and runner for cuda registry#25197

Merged
hnyls2002 merged 6 commits into
mainfrom
lsyin/decouple-stage-runner
May 14, 2026
Merged

ci: decouple stage and runner for cuda registry#25197
hnyls2002 merged 6 commits into
mainfrom
lsyin/decouple-stage-runner

Conversation

@hnyls2002
Copy link
Copy Markdown
Collaborator

Split register_cuda_ci(..., suite="stage-X-test-Y", ...) into two independent fields stage="stage-X" and runner="Y". Schema in ci_register.py accepts either style — suite= is preserved for nightly, AMD, CPU, NPU, and weekly/stress/full-* legacy buckets.

CIRegistry gains an effective_suite property that auto-derives f"{stage}-test-{runner}" when the pair is set, otherwise returns suite. All downstream consumers (run_suite.py:filter_tests, scripts/ci/utils/compute_partitions.py) updated to read effective_suite; workflow yaml unchanged.

Migration scope: 258 register_cuda_ci calls across 256 files. Only suite names matching ^stage-[abc]-test- are split; nightly-, stress, weekly- stay on suite=.

Mechanical refactor + verification script: https://gist.github.com/hnyls2002/932b17106b850029ff4170c539b07c4a

Verified by parsing the registry with the AST parser before and after migration: 954 CIRegistry entries identical modulo effective_suite reconstruction. compute_partitions.py JSON output byte-identical before/after.

@hnyls2002 hnyls2002 requested a review from sundar24295s as a code owner May 13, 2026 23:34
@gemini-code-assist
Copy link
Copy Markdown
Contributor

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

@github-actions github-actions Bot added quant LLM Quantization lora Multi-modal multi-modal language model deepseek speculative-decoding hicache Hierarchical Caching for SGLang blackwell SM100/SM120 npu labels May 13, 2026
@hnyls2002
Copy link
Copy Markdown
Collaborator Author

/tag-and-rerun-ci

@github-actions github-actions Bot added the documentation Improvements or additions to documentation label May 14, 2026
@hnyls2002 hnyls2002 merged commit 22d3f39 into main May 14, 2026
46 of 143 checks passed
@hnyls2002 hnyls2002 deleted the lsyin/decouple-stage-runner branch May 14, 2026 00:28
kpham-sgl added a commit that referenced this pull request May 17, 2026
Per #25197, stage-shaped per-commit CUDA suites must register via
stage=/runner_config= kwargs; the legacy suite= form is reserved for
nightly/stress/weekly and non-stage backends.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

blackwell SM100/SM120 bypass-fastfail deepseek documentation Improvements or additions to documentation hicache Hierarchical Caching for SGLang lora Multi-modal multi-modal language model npu quant LLM Quantization run-ci speculative-decoding

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant