patch_for_mlperf_trining_v1.1_by_samsung.patch

From 15632b0762b52aa12e175045096bd7f6ea756e3d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=EB=AA=85=EC=9A=B0=20=EA=B9=80?= <k.myeong-woo@samsung.com>
Date: Fri, 22 Oct 2021 18:16:53 +0900
Subject: [PATCH] Modify detach_() to detach() to enable Pytorch DDP bucket
 view
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Signed-off-by: 명우 김 <k.myeong-woo@samsung.com>
---
 apex/amp/_process_optimizer.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/apex/amp/_process_optimizer.py b/apex/amp/_process_optimizer.py
index 471289b..b6369d5 100644
--- a/apex/amp/_process_optimizer.py
+++ b/apex/amp/_process_optimizer.py
@@ -370,11 +370,11 @@ def _process_optimizer(optimizer, properties):
             # Zero the model grads.
             for param in stash.all_fp16_params:
                 if param.grad is not None:
-                    param.grad.detach_()
+                    param.grad.detach()
                     param.grad.zero_()
             for param in stash.all_fp32_from_fp32_params:
                 if param.grad is not None:
-                    param.grad.detach_()
+                    param.grad.detach()
                     param.grad.zero_()
             # Clear the master grads that are independent of model grads
             for param in self._amp_stash.all_fp32_from_fp16_params:
-- 
2.17.1