[Distributed] Add custom allreduce support for ROCM (vllm-project#14125)

Signed-off-by: ilmarkov <[email protected]>
Co-authored-by: ilmarkov <[email protected]>
Signed-off-by: Bill Nell <[email protected]>

Author: 2 people

csrc/custom_all_reduce.cuh

@@ -602,4 +602,4 @@ class CustomAllreduce {
  * template void vllm::CustomAllreduce::allreduce<half>(cudaStream_t, half *,
  half *, int, int, int);
 */
-}  // namespace vllm
+}  // namespace vllm