minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

move runs_on + rdma into runner_configs.yml (#25264)

Browse Source
This commit is contained in:
Liangsheng Yin
2026-05-15 04:19:13 -07:00
committed by GitHub
parent 293027aafa
commit fd9525436b
9 changed files with 104 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/ci/runner_configs.py
+43 -10
View File
@@ -1,14 +1,24 @@
"""Emit a runner_config's setup details (install / artifact_version /
install_timeout) in $GITHUB_OUTPUT format. Reads scripts/ci/runner_configs.yml.
Called by .github/workflows/_pr-test-stage.yml.
"""Emit runner_config setup for GitHub Actions $GITHUB_OUTPUT.
runner_configs.py <runner_config>
Per-field `key=value` lines (install / artifact_version /
install_timeout / rdma_devices). `runs_on` is intentionally omitted —
it carries the `$b200_runner` sentinel and is resolved via --map.
Called per stage by _pr-test-stage.yml.
runner_configs.py --map <b200_runner_label>
`runs_on_map={json}` — flat dict {runner_config: runs_on}, with
`$b200_runner` substituted. Called once by _pr-test-check-changes.yml.
"""
import json
import os
import sys
import yaml
_YAML_PATH = os.path.join(os.path.dirname(__file__), "runner_configs.yml")
_B200_SENTINEL = "$b200_runner"
def load() -> dict:
@@ -16,12 +26,35 @@ def load() -> dict:
return yaml.safe_load(f)["runner_configs"]
if __name__ == "__main__":
if len(sys.argv) != 2:
sys.exit("usage: runner_configs.py <runner_config>")
rc = sys.argv[1]
config = load().get(rc)
if config is None:
def _emit_single(rc: str) -> None:
# runs_on goes through --map (resolves $b200_runner). Suppress it here so a
# consumer can't accidentally read the raw sentinel value.
cfg = load().get(rc)
if cfg is None:
sys.exit(f"unknown runner_config: {rc!r}")
for key, value in config.items():
for key, value in cfg.items():
if key == "runs_on":
continue
print(f"{key}={value}")
def _emit_map(b200_runner: str) -> None:
runs_on = {
name: (b200_runner if cfg.get("runs_on") == _B200_SENTINEL else cfg["runs_on"])
for name, cfg in load().items()
}
print(f"runs_on_map={json.dumps(runs_on, separators=(',', ':'))}")
if __name__ == "__main__":
args = sys.argv[1:]
if len(args) == 1:
_emit_single(args[0])
elif len(args) == 2 and args[0] == "--map":
_emit_map(args[1])
else:
sys.exit(
"usage:\n"
" runner_configs.py <runner_config>\n"
" runner_configs.py --map <b200_runner_label>"
)
scripts/ci/runner_configs.yml
+21 -14
View File
@@ -3,9 +3,16 @@
# scripts/ci/runner_configs.py (CLI wrapper), which is in turn called by
# .github/workflows/_pr-test-stage.yml.
#
# Each runner_config carries install script, actions/download-artifact major
# version, and install-step wall-clock cap (minutes, enforced via
# `timeout-minutes:` on the install step in _pr-test-stage.yml).
# Each runner_config carries:
# - install: install script path
# - artifact_version: actions/download-artifact major version
# - install_timeout: install-step wall-clock cap (minutes), enforced via
# `timeout-minutes:` on the install step in _pr-test-stage.yml
# - runs_on: GHA runner label for the stage's `runs-on:`. The literal
# `$b200_runner` is substituted at workflow-load time with the dynamic
# b200 runner tag from check-changes (see runner_configs.py --map).
# - rdma_devices (optional): exported as SGLANG_CI_RDMA_ALL_DEVICES env
# to the stage job; absent means unset.
_anchors:
default_install: &default scripts/ci/cuda/ci_install_dependency.sh
@@ -13,14 +20,14 @@ _anchors:
dsv4_install: &dsv4 scripts/ci/cuda/ci_install_dsv4_dep.sh
runner_configs:
1-gpu-small: { install: *default, artifact_version: v4, install_timeout: "20" }
1-gpu-large: { install: *default, artifact_version: v4, install_timeout: "20" }
2-gpu-large: { install: *default, artifact_version: v4, install_timeout: "20" }
4-gpu-b200: { install: *default, artifact_version: v6, install_timeout: "20" }
4-gpu-h100: { install: *default, artifact_version: v4, install_timeout: "20" }
8-gpu-h200: { install: *default, artifact_version: v4, install_timeout: "20" }
8-gpu-h20: { install: *deepep, artifact_version: v4, install_timeout: "20" }
deepep-4-gpu-h100: { install: *deepep, artifact_version: v4, install_timeout: "20" }
deepep-8-gpu-h200: { install: *deepep, artifact_version: v4, install_timeout: "20" }
dsv4-4-gpu-b200: { install: *dsv4, artifact_version: v6, install_timeout: "30" }
dsv4-8-gpu-h200: { install: *dsv4, artifact_version: v4, install_timeout: "30" }
1-gpu-small: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 1-gpu-5090 }
1-gpu-large: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 1-gpu-h100 }
2-gpu-large: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 2-gpu-h100 }
4-gpu-b200: { install: *default, artifact_version: v6, install_timeout: "20", runs_on: $b200_runner }
4-gpu-h100: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 4-gpu-h100 }
8-gpu-h200: { install: *default, artifact_version: v4, install_timeout: "20", runs_on: 8-gpu-h200 }
8-gpu-h20: { install: *deepep, artifact_version: v4, install_timeout: "20", runs_on: 8-gpu-h20, rdma_devices: "mlx5_1,mlx5_2,mlx5_3,mlx5_4" }
deepep-4-gpu-h100: { install: *deepep, artifact_version: v4, install_timeout: "20", runs_on: 4-gpu-h100 }
deepep-8-gpu-h200: { install: *deepep, artifact_version: v4, install_timeout: "20", runs_on: 8-gpu-h200-deepep }
dsv4-4-gpu-b200: { install: *dsv4, artifact_version: v6, install_timeout: "30", runs_on: $b200_runner }
dsv4-8-gpu-h200: { install: *dsv4, artifact_version: v4, install_timeout: "30", runs_on: 8-gpu-h200 }