[AMD][CI] Consolidate AMD workflows and retire ROCm 7.0 CI (#38632)

This commit is contained in:
Bingxu Chen
2026-09-15 23:45:22 +08:00
committed by GitHub
parent 47a157f257
commit 03ea13a545
17 changed files with 1190 additions and 4932 deletions
+6 -12
View File
@@ -19,8 +19,9 @@ else
fi
# Default base tags (can be overridden by command line arguments)
ROCM_VERSION="rocm700"
# Default base tags (can be overridden by command line arguments).
# Keep this aligned with the default AMD PR and Nightly CI lane.
ROCM_VERSION="rocm10"
DEFAULT_MI30X_BASE_TAG="${SGLANG_VERSION}-${ROCM_VERSION}-mi30x"
DEFAULT_MI35X_BASE_TAG="${SGLANG_VERSION}-${ROCM_VERSION}-mi35x"
LOCAL_DOCKER_REGISTRY="10.44.14.109:5000"
@@ -52,8 +53,8 @@ while [[ $# -gt 0 ]]; do
echo " --mi35x-base-tag TAG Override MI35x base image tag"
echo " --custom-image IMAGE Use a specific Docker image directly (or set AMD_CI_IMAGE)"
echo " --build-from-dockerfile Build image from docker/rocm.Dockerfile"
echo " --gpu-arch ARCH GPU architecture for Dockerfile build (e.g., gfx950-rocm720)"
echo " --rocm-version VERSION Override ROCm version for image lookup (e.g., rocm720)"
echo " --gpu-arch ARCH GPU architecture for Dockerfile build (e.g., gfx950-rocm1000)"
echo " --rocm-version VERSION Override ROCm version for image lookup (e.g., rocm10)"
echo ""
echo "Environment:"
echo " ENABLE_CACHE_HOST=1|0"
@@ -211,13 +212,6 @@ find_latest_image() {
echo "rocm/sgl-dev:v0.5.8.post1-rocm720-mi30x-20260211-preview"
fi
;;
rocm700)
if [[ "${gpu_arch}" == "mi35x" ]]; then
echo "rocm/sgl-dev:v0.5.8.post1-rocm700-mi35x-20260211"
else
echo "rocm/sgl-dev:v0.5.8.post1-rocm700-mi30x-20260211"
fi
;;
*)
echo "Error: no hard-coded fallback available for ${ROCM_VERSION}" >&2
return 1
@@ -253,7 +247,7 @@ elif [[ -n "${BUILD_FROM_DOCKERFILE}" ]]; then
IMAGE="sglang-ci:${GPU_ARCH_BUILD}-$(date +%Y%m%d)"
echo "Building Docker image from ${DOCKERFILE} with GPU_ARCH=${GPU_ARCH_BUILD}..."
# Pass full GPU_ARCH (e.g., gfx950-rocm720) - Dockerfile handles stripping suffix
# Pass full GPU_ARCH (e.g., gfx950-rocm1000) - Dockerfile handles stripping suffix
docker build \
--build-arg GPU_ARCH="${GPU_ARCH_BUILD}" \
--build-arg SGL_BRANCH="main" \
@@ -19,8 +19,9 @@ else
fi
# Default base tags (can be overridden by command line arguments)
ROCM_VERSION="rocm700"
# Default base tags (can be overridden by command line arguments).
# Keep this aligned with the default AMD PR and Nightly CI lane.
ROCM_VERSION="rocm10"
DEFAULT_MI30X_BASE_TAG="${SGLANG_VERSION}-${ROCM_VERSION}-mi30x"
DEFAULT_MI35X_BASE_TAG="${SGLANG_VERSION}-${ROCM_VERSION}-mi35x"
LOCAL_DOCKER_REGISTRY="10.44.14.109:5000"
@@ -196,13 +197,6 @@ find_latest_image() {
echo "rocm/sgl-dev:v0.5.8.post1-rocm720-mi30x-20260211-preview"
fi
;;
rocm700)
if [[ "${gpu_arch}" == "mi35x" ]]; then
echo "rocm/sgl-dev:v0.5.8.post1-rocm700-mi35x-20260211"
else
echo "rocm/sgl-dev:v0.5.8.post1-rocm700-mi30x-20260211"
fi
;;
*)
echo "Error: no hard-coded fallback available for ${ROCM_VERSION}" >&2
return 1