diff --git a/PaddleCV/video/models/tsm/README.md b/PaddleCV/video/models/tsm/README.md
index 0fe81f7d1019f86988366eea15b10d514b82fa0c..a92945104c65fc7bcfa7dcac5b3b0485a44a194b 100644
--- a/PaddleCV/video/models/tsm/README.md
+++ b/PaddleCV/video/models/tsm/README.md
@@ -34,6 +34,9 @@ TSM的训练数据采用由DeepMind公布的Kinetics-400动作识别数据集。
 
 数据准备完毕后，可以通过如下两种方式启动训练：
 
+    export FLAGS_fast_eager_deletion_mode=1
+    export FLAGS_eager_delete_tensor_gb=0.0
+    export FLAGS_fraction_of_gpu_memory_to_use=0.98
     python train.py --model_name=TSM
             --config=./configs/tsm.txt
             --save_dir=checkpoints
diff --git a/PaddleCV/video/scripts/train/train_tsm.sh b/PaddleCV/video/scripts/train/train_tsm.sh
index d09096637211f70889d00b5a68f1dcd5df010cc2..dcab59f79dc4ed987683987f1e3da14ad497b56f 100644
--- a/PaddleCV/video/scripts/train/train_tsm.sh
+++ b/PaddleCV/video/scripts/train/train_tsm.sh
@@ -1,3 +1,9 @@
 export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
+
+# activate eager gc to reduce memory use
+export FLAGS_fast_eager_deletion_mode=1
+export FLAGS_eager_delete_tensor_gb=0.0
+export FLAGS_fraction_of_gpu_memory_to_use=0.98
+
 python train.py --model_name="TSM" --config=./configs/tsm.txt --epoch_num=65 \
                 --valid_interval=1 --log_interval=10