[Core] Remove the legacy Ray compiled-graph executor (#61712)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5.5
parent
bfdbfa1632
commit
e57bb4ace7
+7
-30
@@ -60,11 +60,7 @@ if TYPE_CHECKING:
|
||||
VLLM_XLA_CHECK_RECOMPILATION: bool = False
|
||||
VLLM_SPARSE_INDEXER_MAX_LOGITS_MB: int = 512
|
||||
VLLM_ADAPTIVE_VERIFICATION_PROFILE_CONTEXT_LEN: int = 8192
|
||||
VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE: Literal["auto", "nccl", "shm"] = "auto"
|
||||
VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM: bool = False
|
||||
VLLM_USE_RAY_WRAPPED_PP_COMM: bool = True
|
||||
VLLM_XPU_PP_MICROBATCH: bool = True
|
||||
VLLM_USE_RAY_V2_EXECUTOR_BACKEND: bool = False
|
||||
VLLM_DISTRIBUTED_USE_SPLIT_GROUP: bool = False
|
||||
VLLM_XLA_USE_SPMD: bool = False
|
||||
VLLM_WORKER_MULTIPROC_METHOD: Literal["fork", "spawn"] = "fork"
|
||||
@@ -174,6 +170,7 @@ if TYPE_CHECKING:
|
||||
VLLM_SERVER_DEV_MODE: bool = False
|
||||
VLLM_V1_OUTPUT_PROC_CHUNK_SIZE: int = 128
|
||||
VLLM_MLA_DISABLE: bool = False
|
||||
VLLM_USE_RAY_V2_EXECUTOR_BACKEND: bool = True
|
||||
VLLM_RAY_PER_WORKER_GPUS: float = 1.0
|
||||
VLLM_RAY_BUNDLE_INDICES: str = ""
|
||||
VLLM_CUDART_SO_PATH: str | None = None
|
||||
@@ -914,37 +911,11 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
# (CPU backend only) whether to use SGLang INT4 W4A8 kernels for AWQ, and
|
||||
# on Zen CPUs whether to serve int4 checkpoints as DA8W4 rather than W4A16.
|
||||
"VLLM_CPU_INT4_W4A8": lambda: bool(int(os.getenv("VLLM_CPU_INT4_W4A8", "1"))),
|
||||
# If the env var is set, Ray Compiled Graph uses the specified
|
||||
# channel type to communicate between workers belonging to
|
||||
# different pipeline-parallel stages.
|
||||
# Available options:
|
||||
# - "auto": use the default channel type
|
||||
# - "nccl": use NCCL for communication
|
||||
# - "shm": use shared memory and gRPC for communication
|
||||
"VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE": env_with_choices(
|
||||
"VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE", "auto", ["auto", "nccl", "shm"]
|
||||
),
|
||||
# If the env var is set, it enables GPU communication overlap
|
||||
# (experimental feature) in Ray's Compiled Graph.
|
||||
"VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM", "0"))
|
||||
),
|
||||
# If the env var is set, it uses a Ray Communicator wrapping
|
||||
# vLLM's pipeline parallelism communicator to interact with Ray's
|
||||
# Compiled Graph. Otherwise, it uses Ray's NCCL communicator.
|
||||
"VLLM_USE_RAY_WRAPPED_PP_COMM": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_WRAPPED_PP_COMM", "1"))
|
||||
),
|
||||
# Using a flag to control microbatch on XPU device, users on XPU device can
|
||||
# decide to disable it when they needed.
|
||||
"VLLM_XPU_PP_MICROBATCH": lambda: bool(
|
||||
int(os.getenv("VLLM_XPU_PP_MICROBATCH", "1"))
|
||||
),
|
||||
# When True and distributed_executor_backend="ray", use RayExecutorV2
|
||||
# (MQ-based) instead of RayDistributedExecutor (compiled-graph backend).
|
||||
"VLLM_USE_RAY_V2_EXECUTOR_BACKEND": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_V2_EXECUTOR_BACKEND", "1"))
|
||||
),
|
||||
# When True, GroupCoordinator constructs its CPU/device subgroups via
|
||||
# ``torch.distributed.split_group(backend=...)``
|
||||
# and ``init_distributed_environment`` initializes the default PG with
|
||||
@@ -1513,6 +1484,12 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
# which indices are used for the Ray bundle, for every worker.
|
||||
# Format: comma-separated list of integers, e.g. "0,1,2,3"
|
||||
"VLLM_RAY_BUNDLE_INDICES": lambda: os.getenv("VLLM_RAY_BUNDLE_INDICES", ""),
|
||||
# Deprecated and ignored: RayExecutorV2 is the only Ray executor. Kept so
|
||||
# out-of-tree platforms that still read it keep working; will be removed
|
||||
# in v0.34.
|
||||
"VLLM_USE_RAY_V2_EXECUTOR_BACKEND": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_V2_EXECUTOR_BACKEND", "1"))
|
||||
),
|
||||
# In some system, find_loaded_library() may not work. So we allow users to
|
||||
# specify the path through environment variable VLLM_CUDART_SO_PATH.
|
||||
"VLLM_CUDART_SO_PATH": lambda: os.getenv("VLLM_CUDART_SO_PATH", None),
|
||||
|
||||
Reference in New Issue
Block a user