Fix `dist.broadcast` stall without group argument (#3408)

429284dc · Junda Chen · GitHub · 253a9807 · 429284dc
Unverified Commit 429284dc authored Mar 14, 2024 by Junda Chen Committed by GitHub Mar 14, 2024
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

vllm/model_executor/parallel_utils/communication_op.py vllm/model_executor/parallel_utils/communication_op.py +1 -1

No files found.
--- a/vllm/model_executor/parallel_utils/communication_op.py
+++ b/vllm/model_executor/parallel_utils/communication_op.py
@@ -177,7 +177,7 @@ def broadcast_tensor_dict(
        for key, value in metadata_list:
            if isinstance(value, TensorMetadata):
                tensor = tensor_dict[key]
-                torch.distributed.broadcast(tensor, src=src)
+                torch.distributed.broadcast(tensor, src=src, group=group)
    else:
        recv_metadata_list = [None]
        torch.distributed.broadcast_object_list(recv_metadata_list,