fix pp memory issue (#6344)

TongLi3701 · Tong Li · web-flow · commit 8992def757f7 · 2025-06-11T17:54:18.000+08:00
Co-authored-by: Tong Li &lt;tong.li35271158@gmail.com&gt;
diff --git a/applications/ColossalChat/coati/distributed/grpo_consumer.py b/applications/ColossalChat/coati/distributed/grpo_consumer.py
@@ -345,7 +345,7 @@ def _criterion(outputs, inputs):
                         criterion=_criterion,
                         optimizer=self.optimizer,
                         return_loss=True,
-                        return_outputs=True,
+                        return_outputs=False,
                     )
                     loss = policy_model_outputs["loss"]
 

Original file line number	Diff line number	Diff line change
`@@ -345,7 +345,7 @@ def _criterion(outputs, inputs):`
`345`	`345`	`criterion=_criterion,`
`346`	`346`	`optimizer=self.optimizer,`
`347`	`347`	`return_loss=True,`
`348`		`- return_outputs=True,`
	`348`	`+ return_outputs=False,`
`349`	`349`	`)`
`350`	`350`	`loss = policy_model_outputs["loss"]`
`351`	`351`