From c4c0d08782dacf183c52ce3faf48c352ff3481d8 Mon Sep 17 00:00:00 2001
From: Woosuk Kwon <woosuk.kwon@berkeley.edu>
Date: Tue, 30 Jan 2024 14:39:40 -0800
Subject: [PATCH] [Minor] Fix false warning when TP=1 (#2674)

---
 vllm/model_executor/parallel_utils/custom_all_reduce.py | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/vllm/model_executor/parallel_utils/custom_all_reduce.py b/vllm/model_executor/parallel_utils/custom_all_reduce.py
index 5b88649cc2129..628c151761fb2 100644
--- a/vllm/model_executor/parallel_utils/custom_all_reduce.py
+++ b/vllm/model_executor/parallel_utils/custom_all_reduce.py
@@ -29,6 +29,10 @@ def init_custom_ar() -> None:
         return
     rank = get_tensor_model_parallel_rank()
     world_size = get_tensor_model_parallel_world_size()
+    if world_size == 1:
+        # No need to initialize custom allreduce for single GPU case.
+        return
+
     if world_size not in _SUPPORTED_WORLD_SIZES:
         logger.warn(
             "Custom allreduce is disabled due to an unsupported world size: "