update custom_all_reduce

vllm/config.py

@@ -1411,7 +1411,7 @@ class ParallelConfig:
         if self.use_ray:
             from vllm.executor import ray_utils
             ray_utils.assert_ray_available()
-        # xiabo
+        
         # if current_platform.is_rocm():
         #     self.disable_custom_all_reduce = True
         #     logger.info(

vllm/distributed/device_communicators/custom_all_reduce.py

@@ -136,7 +136,7 @@ class CustomAllreduce:
         assert current_platform.is_cuda_alike()
         full_nvlink = current_platform.is_fully_connected_nvlink_or_xgmi(
             physical_device_ids)
-        if world_size > 2 and not full_nvlink:
+        if not full_nvlink:
             logger.warning(
                 "Custom allreduce is disabled because it's not supported on"
                 " more than two PCIe-only GPUs. To silence this warning, "