[Bugfix] Don't reuse engine core payload buffer while zmq is sending it (#50053)

Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
This commit is contained in:
Nick Hill
2026-07-28 16:09:28 +00:00
committed by GitHub
co-authored by Andreas Karatzas
parent 4fb483ca86
commit 6453fc0b8c
4 changed files with 179 additions and 58 deletions
+1 -1
View File
@@ -1547,7 +1547,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
# tensors above will instead be sent via a separate message.
# While the sending side still actually copies the tensor
# in all cases, on the receiving side, tensors above this
# limit will actually be zero-copy decoded.
# limit will actually be zero-copy decoded. The unit is bytes.
"VLLM_MSGPACK_ZERO_COPY_THRESHOLD": lambda: int(
os.getenv("VLLM_MSGPACK_ZERO_COPY_THRESHOLD", "256")
),