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

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
+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 }