diff --git a/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.md b/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.osc.md
similarity index 96%
rename from PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.md
rename to PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.osc.md
index fca3b25fe89f9d769d93dc0330effa293487e2ad..1bd5da688c2498d4d2773b790d6e75893605261c 100644
--- a/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.md
+++ b/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/README.osc.md
@@ -151,6 +151,7 @@ bash test/train_grpo_performance_16p.sh --model_path=./models/xxx --dataset_path
 # FAQ
 
 * 使用--adam_offload参数可能存在长时间卡顿的情况，解决方法是删除torch_extensions的缓存文件，参考[issue](https://github.com/deepspeedai/DeepSpeed/issues/2816#issuecomment-1450095538)。
+* 在 Atlas 200T A2 Box16 机器中，如果使用了跨平面的卡，需要使能环境变量 `export HCCL_INTRA_ROCE_ENABLE=1`，使用RoCE环路进行多卡间的通信。
 
 # 公网地址说明
 
diff --git a/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/setup.py b/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/setup.py
index 78f255874489c5269d10608af5e77af2c50da75e..44425d3c9c97779ccc1e3f0e89c1a8be63e0e3bd 100644
--- a/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/setup.py
+++ b/PyTorch/built-in/rl/OpenRLHF_v0.6.2_for_PyTorch/setup.py
@@ -19,7 +19,7 @@ def _fetch_requirements(path):
 
 
 def _fetch_readme():
-    with open("README.md", encoding="utf-8") as f:
+    with open("README.osc.md", encoding="utf-8") as f:
         return f.read()