From 4eb17b26e77611d4fbcdcbbc20c7bf275eb015c9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Quentin=20Gallou=C3=A9dec?=
 <45557362+qgallouedec@users.noreply.github.com>
Date: Thu, 26 Dec 2024 14:58:53 +0100
Subject: [PATCH] Drop inplace operation for loss computation with gradient
 accumulation (#35416)

Fix inplace loss computation
---
 src/transformers/trainer.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
index 5957f8025d2..c2327739549 100755
--- a/src/transformers/trainer.py
+++ b/src/transformers/trainer.py
@@ -3700,7 +3700,7 @@ def training_step(
         else:
             # Finally we need to normalize the loss for reporting
             if num_items_in_batch is None:
-                loss /= self.args.gradient_accumulation_steps
+                loss = loss / self.args.gradient_accumulation_steps
 
             self.accelerator.backward(loss, **kwargs)