xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 00:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix V1 engine serialization error with Ray distributed executor (#26148)

Browse Source 
Signed-off-by: Nikhil Ghosh <nikhil@anyscale.com>
This commit is contained in:
Nikhil G 2025-10-03 11:39:45 -07:00 committed by GitHub
parent 300a59c4c3
commit cd9e5b8340
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/executor/ray_utils.py
View File

@ -16,6 +16,7 @@ from vllm.logger import init_logger
from vllm.platforms import current_platform from vllm.platforms import current_platform
from vllm.sequence import ExecuteModelRequest, IntermediateTensors from vllm.sequence import ExecuteModelRequest, IntermediateTensors
from vllm.utils import get_ip from vllm.utils import get_ip
from vllm.v1.outputs import AsyncModelRunnerOutput
from vllm.v1.worker.worker_base import WorkerWrapperBase from vllm.v1.worker.worker_base import WorkerWrapperBase
if TYPE_CHECKING: if TYPE_CHECKING:
@ -142,6 +143,11 @@ try:
# but may still be finished requests. # but may still be finished requests.
assert not output or not output.req_ids assert not output or not output.req_ids
output = scheduler_output, None output = scheduler_output, None
# Ensure outputs crossing Ray compiled DAG are serializable.
# AsyncModelRunnerOutput holds CUDA events and cannot be
# pickled.
if isinstance(output, AsyncModelRunnerOutput):
output = output.get_output()
return output return output
def override_env_vars(self, vars: Dict[str, str]): def override_env_vars(self, vars: Dict[str, str]):