A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 22:35 UTC
[Misc] update doc comment for send (#22026)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Aug 3, 2025 at 15:55 UTC
7de45db9a5b95073c3f99eec75ae510d347d625f
3 files changed
+3
-3
vllm/distributed/device_communicators/base_device_communicator.py
+1
-1
@@ -219,7 +219,7 @@ class DeviceCommunicatorBase:
219
return output_tensor
220
221
def send(self, tensor: torch.Tensor, dst: Optional[int] = None) -> None:
222
- """Sends a tensor to the destination rank in a non-blocking way"""
222
+ """Sends a tensor to the destination rank in a blocking way"""
223
"""NOTE: `dst` is the local rank of the destination rank."""
224
if dst is None:
225
dst = (self.rank_in_group + 1) % self.world_size
vllm/distributed/device_communicators/cuda_communicator.py
+1
-1
@@ -179,7 +179,7 @@ class CudaCommunicator(DeviceCommunicatorBase):
179
return output.movedim(0, dim).contiguous()
180
181
def send(self, tensor: torch.Tensor, dst: Optional[int] = None) -> None:
182
- """Sends a tensor to the destination rank in a non-blocking way"""
182
+ """Sends a tensor to the destination rank in a blocking way"""
183
"""NOTE: `dst` is the local rank of the destination rank."""
184
if dst is None:
185
dst = (self.rank_in_group + 1) % self.world_size
vllm/distributed/parallel_state.py
+1
-1
@@ -782,7 +782,7 @@ class GroupCoordinator:
782
torch.distributed.barrier(group=self.cpu_group)
783
784
def send(self, tensor: torch.Tensor, dst: Optional[int] = None) -> None:
785
- """Sends a tensor to the destination rank in a non-blocking way"""
785
+ """Sends a tensor to the destination rank in a blocking way"""
786
"""NOTE: `dst` is the local rank of the destination rank."""
787
self.device_communicator.send(tensor, dst)
788