[docker] Build CUDA kernels in separate Docker stage for faster rebuilds (#29452)

Signed-off-by: Amr Mahdi <amrmahdi@meta.com>
This commit is contained in:
Amr Mahdi
2025-12-03 11:41:53 +00:00
committed by GitHub
parent 78f4bb0ba8
commit f5d3d93c40
4 changed files with 74 additions and 11 deletions
+5
View File
@@ -78,6 +78,7 @@ if TYPE_CHECKING:
MAX_JOBS: str | None = None
NVCC_THREADS: str | None = None
VLLM_USE_PRECOMPILED: bool = False
VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX: bool = False
VLLM_DOCKER_BUILD_CONTEXT: bool = False
VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL: bool = False
VLLM_KEEP_ALIVE_ON_ENGINE_DEATH: bool = False
@@ -462,6 +463,10 @@ environment_variables: dict[str, Callable[[], Any]] = {
.lower()
in ("1", "true")
or bool(os.environ.get("VLLM_PRECOMPILED_WHEEL_LOCATION")),
# If set, skip adding +precompiled suffix to version string
"VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX": lambda: bool(
int(os.environ.get("VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX", "0"))
),
# Used to mark that setup.py is running in a Docker build context,
# in order to force the use of precompiled binaries.
"VLLM_DOCKER_BUILD_CONTEXT": lambda: os.environ.get("VLLM_DOCKER_BUILD_CONTEXT", "")