xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-29 10:14:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc] Increase RayDistributedExecutor RAY_CGRAPH_get_timeout (#15301)

Browse Source 
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
This commit is contained in:
Rui Qiao 2025-03-21 22:25:43 -07:00 committed by GitHub
parent 790b79750b
commit 4c69e228b3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 9 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/executor/ray_distributed_executor.py
View File

@ -561,6 +561,15 @@ class RayDistributedExecutor(DistributedExecutorBase):
envs.VLLM_USE_RAY_COMPILED_DAG_NCCL_CHANNEL)
logger.info("VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM = %s",
envs.VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM)
# Enlarge the default value of "RAY_CGRAPH_get_timeout" to 300 seconds
# (it is 10 seconds by default). This is a Ray environment variable to
# control the timeout of getting result from a compiled graph execution,
# i.e., the distributed execution that includes model forward runs and
# intermediate tensor communications, in the case of vllm.
os.environ.setdefault("RAY_CGRAPH_get_timeout", "300") # noqa: SIM112
logger.info("RAY_CGRAPH_get_timeout is set to %s",
os.environ["RAY_CGRAPH_get_timeout"]) # noqa: SIM112
with InputNode() as input_data:
# Example DAG: PP=2, TP=4
#