From ca6a3bc76f070104d62ac11f30548d5433aff9b0 Mon Sep 17 00:00:00 2001
From: hiyouga <hiyouga@buaa.edu.cn>
Date: Wed, 27 Sep 2023 22:49:43 +0800
Subject: [PATCH] fix #424

Former-commit-id: 35fa94723c7d0c024a63f47208291d09c67fc13e
---
 src/llmtuner/tuner/ppo/workflow.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/llmtuner/tuner/ppo/workflow.py b/src/llmtuner/tuner/ppo/workflow.py
index be8fca5d..ed00b437 100644
--- a/src/llmtuner/tuner/ppo/workflow.py
+++ b/src/llmtuner/tuner/ppo/workflow.py
@@ -42,7 +42,8 @@ def run_ppo(
         ppo_epochs=1,
         max_grad_norm=training_args.max_grad_norm,
         seed=training_args.seed,
-        optimize_cuda_cache=True
+        optimize_cuda_cache=True,
+        accelerator_kwargs={"step_scheduler_with_optimizer": False}
     )
 
     if finetuning_args.ppo_score_norm: