frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

update

Browse Source 
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
This commit is contained in:
Tyler Michael Smith
2025-07-10 18:27:59 -04:00
parent 29596317b0
commit df866cfebf
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/distributed/parallel_state.py
View File

@ -782,7 +782,6 @@ class GroupCoordinator:
def prepare_communication_buffer_for_model(self, model: torch.nn.Module):
if self.device_communicator is not None:
torch.distributed.barrier(self.device_communicator)
self.device_communicator.prepare_communication_buffer_for_model(
model)
@ -1116,6 +1115,7 @@ def prepare_communication_buffer_for_model(model: torch.nn.Module):
MoE all2all (DeepEP) usually allocate the communication buffer
based on the model shape for optimal performance.
"""
get_world_group().barrier()
if _TP is not None:
_TP.prepare_communication_buffer_for_model(model)
if _PP is not None: