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

[AMD] Replace MI325 with MI300 CI Runners (#31409)

Browse Source
This commit is contained in:
Bingxu Chen
2026-07-29 08:18:07 -07:00
committed by GitHub
parent e1f2f9d1fa
commit bfc450248e
9 changed files with 200 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/ci/amd/amd_ci_start_container.sh
+17 -9
View File
@@ -57,7 +57,7 @@ while [[ $# -gt 0 ]]; do
echo ""
echo "Environment:"
echo " ENABLE_CACHE_HOST=1|0"
echo " Mount /home/runner/sglang-data to /sgl-data. Defaults to 0."
echo " Mount /home/runner/sglang-data to /sgl-data. Defaults to 1 when RUNNER_NAME contains 300 or 35x, otherwise 0. Missing host cache falls back to container-local /sgl-data."
exit 0
;;
*) echo "Unknown option $1"; exit 1;;
@@ -288,15 +288,24 @@ else
fi
CACHE_HOST=/home/runner/sglang-data
ENABLE_CACHE_HOST="${ENABLE_CACHE_HOST:-0}"
if [[ -z "${ENABLE_CACHE_HOST:-}" ]]; then
RUNNER_NAME_LOWER="${RUNNER_NAME:-}"
RUNNER_NAME_LOWER="${RUNNER_NAME_LOWER,,}"
if [[ "${RUNNER_NAME_LOWER}" == *300* || "${RUNNER_NAME_LOWER}" == *35x* ]]; then
ENABLE_CACHE_HOST="1"
else
ENABLE_CACHE_HOST="0"
fi
fi
case "${ENABLE_CACHE_HOST,,}" in
1|true|yes|on|pvc|persistent)
if [[ ! -d "$CACHE_HOST" ]]; then
echo "Error: ENABLE_CACHE_HOST=1 but ${CACHE_HOST} does not exist." >&2
exit 1
if [[ -d "$CACHE_HOST" ]]; then
CACHE_VOLUME="-v $CACHE_HOST:/sgl-data"
echo "Mounting persistent CI data: ${CACHE_HOST} -> /sgl-data"
else
CACHE_VOLUME=""
echo "Warning: ${CACHE_HOST} does not exist; using container-local /sgl-data." >&2
fi
CACHE_VOLUME="-v $CACHE_HOST:/sgl-data"
echo "Mounting persistent CI data: ${CACHE_HOST} -> /sgl-data"
;;
0|false|no|off|"")
CACHE_VOLUME=""
@@ -332,8 +341,7 @@ docker run -dt --user root --device=/dev/kfd ${DEVICE_FLAG} \
docker exec ci_sglang mkdir -p \
/sgl-data/hf-cache/hub \
/sgl-data/pip-cache \
/sgl-data/miopen-cache \
/sgl-data/aiter-kernels
/sgl-data/miopen-cache
# The checkout is owned by the runner (non-root) but the container runs as
# root. Git >= 2.35.2 rejects cross-user repos; mark the mount as safe so
scripts/ci/amd/amd_ci_start_container_disagg.sh
+34 -6
View File
@@ -228,13 +228,36 @@ else
retry_with_backoff 6 docker pull "${IMAGE}"
fi
# CACHE_HOST=/home/runner/sgl-data
CACHE_HOST=/home/runner/temp-sglang-data
if [[ -d "$CACHE_HOST" ]]; then
CACHE_VOLUME="-v $CACHE_HOST:/sgl-data"
else
CACHE_VOLUME=""
CACHE_HOST=/home/runner/sglang-data
if [[ -z "${ENABLE_CACHE_HOST:-}" ]]; then
RUNNER_NAME_LOWER="${RUNNER_NAME:-}"
RUNNER_NAME_LOWER="${RUNNER_NAME_LOWER,,}"
if [[ "${RUNNER_NAME_LOWER}" == *300* || "${RUNNER_NAME_LOWER}" == *35x* ]]; then
ENABLE_CACHE_HOST="1"
else
ENABLE_CACHE_HOST="0"
fi
fi
case "${ENABLE_CACHE_HOST,,}" in
1|true|yes|on|pvc|persistent)
if [[ -d "$CACHE_HOST" ]]; then
CACHE_VOLUME="-v $CACHE_HOST:/sgl-data"
echo "Mounting persistent CI data: ${CACHE_HOST} -> /sgl-data"
else
CACHE_VOLUME=""
echo "Warning: ${CACHE_HOST} does not exist; using container-local /sgl-data." >&2
fi
;;
0|false|no|off|"")
CACHE_VOLUME=""
echo "Not mounting ${CACHE_HOST}; /sgl-data will be container-local."
;;
*)
echo "Error: unsupported ENABLE_CACHE_HOST='${ENABLE_CACHE_HOST}'" >&2
echo "Use 1/true/pvc/persistent or 0/false/off." >&2
exit 1
;;
esac
# Detect libionic library for RDMA support
LIBIONIC_MOUNT=""
@@ -325,6 +348,11 @@ docker run -dt --user root \
--name ci_sglang \
"${IMAGE}"
docker exec ci_sglang mkdir -p \
/sgl-data/hf-cache/hub \
/sgl-data/pip-cache \
/sgl-data/miopen-cache
# The checkout is owned by the runner (non-root) but the container runs as
# root. Git >= 2.35.2 rejects cross-user repos; mark the mount as safe so
# setuptools-scm / vcs_versioning can resolve the package version.