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 17, 04:15 UTC
Fix `dist.broadcast` stall without group argument (#3408)
Junda Chen committed
Mar 14, 2024 at 23:25 UTC
429284dc374bab79d4dfbb25053583901e6e5051
1 file changed
+1
-1
vllm/model_executor/parallel_utils/communication_op.py
+1
-1
@@ -177,7 +177,7 @@ def broadcast_tensor_dict(
177
for key, value in metadata_list:
178
if isinstance(value, TensorMetadata):
179
tensor = tensor_dict[key]
180
- torch.distributed.broadcast(tensor, src=src)
180
+ torch.distributed.broadcast(tensor, src=src, group=group)
181
else:
182
recv_metadata_list = [None]
183
torch.distributed.broadcast_object_list(recv_metadata_list,