[V0 Deprecation] Remove V0 executors (#27142)
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
+1
-22
@@ -56,8 +56,6 @@ if TYPE_CHECKING:
|
||||
VLLM_XLA_CHECK_RECOMPILATION: bool = False
|
||||
VLLM_FUSED_MOE_CHUNK_SIZE: int = 64 * 1024
|
||||
VLLM_ENABLE_FUSED_MOE_ACTIVATION_CHUNKING: bool = True
|
||||
VLLM_USE_RAY_SPMD_WORKER: bool = False
|
||||
VLLM_USE_RAY_COMPILED_DAG: bool = False
|
||||
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
|
||||
@@ -623,22 +621,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
"VLLM_CPU_MOE_PREPACK": lambda: bool(int(os.getenv("VLLM_CPU_MOE_PREPACK", "1"))),
|
||||
# (CPU backend only) whether to use SGL kernels, optimized for small batch.
|
||||
"VLLM_CPU_SGL_KERNEL": lambda: bool(int(os.getenv("VLLM_CPU_SGL_KERNEL", "0"))),
|
||||
# If the env var is set, then all workers will execute as separate
|
||||
# processes from the engine, and we use the same mechanism to trigger
|
||||
# execution on all workers.
|
||||
# Run vLLM with VLLM_USE_RAY_SPMD_WORKER=1 to enable it.
|
||||
"VLLM_USE_RAY_SPMD_WORKER": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_SPMD_WORKER", "0"))
|
||||
),
|
||||
# If the env var is set, it uses the Ray's Compiled Graph
|
||||
# (previously known as ADAG) API which optimizes the
|
||||
# control plane overhead.
|
||||
# Run vLLM with VLLM_USE_RAY_COMPILED_DAG=1 to enable it.
|
||||
# Note that this variable is set to 1 in V1 by default
|
||||
# when ray distributed executor is used.
|
||||
"VLLM_USE_RAY_COMPILED_DAG": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_COMPILED_DAG", "0"))
|
||||
),
|
||||
# If the env var is set, Ray Compiled Graph uses the specified
|
||||
# channel type to communicate between workers belonging to
|
||||
# different pipeline-parallel stages.
|
||||
@@ -646,20 +628,17 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
# - "auto": use the default channel type
|
||||
# - "nccl": use NCCL for communication
|
||||
# - "shm": use shared memory and gRPC for communication
|
||||
# This flag is ignored if VLLM_USE_RAY_COMPILED_DAG is not set.
|
||||
"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. This flag is ignored if
|
||||
# VLLM_USE_RAY_COMPILED_DAG is not set.
|
||||
# (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.
|
||||
# This flag is ignored if VLLM_USE_RAY_COMPILED_DAG is not set.
|
||||
"VLLM_USE_RAY_WRAPPED_PP_COMM": lambda: bool(
|
||||
int(os.getenv("VLLM_USE_RAY_WRAPPED_PP_COMM", "1"))
|
||||
),
|
||||
|
||||
Reference in New Issue
Block a user