biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[core] Bump ray to use _overlap_gpu_communication in compiled graph tests (#10410)

Browse Source 
Signed-off-by: Rui Qiao <ubuntu@ip-172-31-15-128.us-west-2.compute.internal>
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
Co-authored-by: Rui Qiao <ubuntu@ip-172-31-15-128.us-west-2.compute.internal>
This commit is contained in:
Rui Qiao
2024-12-11 11:36:35 -08:00
committed by GitHub
parent 66aaa7722d
commit 72ff3a9686
4 changed files with 20 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/envs.py
View File

@@ -45,6 +45,7 @@ if TYPE_CHECKING:
VLLM_USE_RAY_SPMD_WORKER: bool = False
VLLM_USE_RAY_COMPILED_DAG: bool = False
VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL: bool = True
VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM: bool = True
VLLM_WORKER_MULTIPROC_METHOD: str = "fork"
VLLM_ASSETS_CACHE: str = os.path.join(VLLM_CACHE_ROOT, "assets")
VLLM_IMAGE_FETCH_TIMEOUT: int = 5
@@ -337,6 +338,13 @@ environment_variables: Dict[str, Callable[[], Any]] = {
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL", "1"))
),
# If the env var is set, it enables GPU communication overlap in
# Ray's compiled DAG. This flag is ignored if
# VLLM_USE_RAY_COMPILED_DAG is not set.
"VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM", "1"))
),
# Use dedicated multiprocess context for workers.
# Both spawn and fork work
"VLLM_WORKER_MULTIPROC_METHOD":