From 30978bea2cef09ac9740e0e07272b0dbebc5fd71 Mon Sep 17 00:00:00 2001
From: linhu-nv <linhu@nvidia.com>
Date: Tue, 5 Mar 2024 18:31:00 +0800
Subject: [PATCH] remove an unnecessary sync in exchange_embeddings_nccl_func

---
 .../functions/exchange_embeddings_nccl_func.cu                | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/cpp/src/wholememory_ops/functions/exchange_embeddings_nccl_func.cu b/cpp/src/wholememory_ops/functions/exchange_embeddings_nccl_func.cu
index 6cb9af49d..7cb96bcb4 100644
--- a/cpp/src/wholememory_ops/functions/exchange_embeddings_nccl_func.cu
+++ b/cpp/src/wholememory_ops/functions/exchange_embeddings_nccl_func.cu
@@ -62,8 +62,8 @@ wholememory_error_code_t exchange_embeddings_nccl_func(const void* dev_local_gat
                        WHOLEMEMORY_DT_INT8,
                        stream);
     WM_CUDA_DEBUG_SYNC_STREAM(stream);
-    WHOLEMEMORY_EXPECTS(wm_comm->sync_stream(stream) == WHOLEMEMORY_SUCCESS,
-                        "Embedding AllToAllV failed.");
+    // WHOLEMEMORY_EXPECTS(wm_comm->sync_stream(stream) == WHOLEMEMORY_SUCCESS,
+    //                    "Embedding AllToAllV failed.");
   } catch (wholememory::logic_error& wle) {
     WHOLEMEMORY_ERROR("exchange_embeddings_nccl_func LOGIC Error %s\n", wle.what());
     return WHOLEMEMORY_LOGIC_ERROR;