diff --git a/README.md b/README.md
index 58b0bd325878ff9c68927f0717ad5ed0688dab41..44c53973857fa962aac72b5387dc69e8274ea232 100644
--- a/README.md
+++ b/README.md
@@ -1009,3 +1009,7 @@ Export the path of cuda to LD_LIBRARY_PATH, e.g.: `export LD_LIBRARY_PATH=/home/
 #### FAQ5: Can not find library: libnccl.so. Please try to add the lib path to LD_LIBRARY_PATH.
 
 Download [NCCL2](https://developer.nvidia.com/nccl/nccl-download), and export the library path to LD_LIBRARY_PATH, e.g.:`export LD_LIBRARY_PATH=/home/work/nccl/lib`
+
+#### FAQ6: Cannot malloc XXX MB GPU memory.
+
+Try to reduce the batch_size, reduce the max_seq_len and set FLAGS_eager_delete_tensor_gb=0.0
diff --git a/README.zh.md b/README.zh.md
index 461042bb69a49213f56011a7aa0d85235c5db99b..b7e6d09061a72cca43e54d3f8e1384f162167e86 100644
--- a/README.zh.md
+++ b/README.zh.md
@@ -1102,3 +1102,7 @@ python -u infer_classifyer.py \
 ### FQA6: 运行报错`ModuleNotFoundError: No module named 'propeller'`<a name="faq6"></a>
 
 您可以通过`export PYTHONPATH=./:$PYTHONPATH`的方式引入Propeller.
+
+### FAQ7：显存不足报错: Cannot malloc XXX MB GPU memory. 
+
+您可以通过减小batch_size, 减小max_seq_len, 设置FLAGS_eager_delete_tensor_gb=0.0来减小网络对显存的需求