diff --git a/docs/tutorial/image_cls.md b/docs/tutorial/image_cls.md
index bffe471b9..b16e0d791 100644
--- a/docs/tutorial/image_cls.md
+++ b/docs/tutorial/image_cls.md
@@ -106,7 +106,7 @@ If configuration finishes, the configuration file is generated in the `cifar10_t
 Train your model by running `blueoil train` with model configuration.
 
     $ docker run --rm \
-        -e CUDA_VISIBLE_DEVICES=0 \
+        --gpus '"device=0"' \
         -v $(pwd)/cifar:/home/blueoil/cifar \
         -v $(pwd)/config:/home/blueoil/config \
         -v $(pwd)/saved:/home/blueoil/saved \
@@ -114,7 +114,8 @@ Train your model by running `blueoil train` with model configuration.
         blueoil train -c config/cifar10_test.py
 
 Just like init, set the value of `{TAG}` to the value obtained by `docker images`.
-Change the value of `CUDA_VISIBLE_DEVICES` according to your environment.
+Change the value of `--gpus '"device=<gpu indices>"'` according to your environment. 
+For detail check the official [documentation](https://docs.docker.com/config/containers/resource_constraints/#gpu).
 
 When training has started, the training log and checkpoints are generated under `./saved/{MODEL_NAME}`.
 The value of `{MODEL_NAME}` will be `{Configuration file}_{TIMESTAMP}`.
diff --git a/docs/tutorial/image_det.md b/docs/tutorial/image_det.md
index 437f14d6c..d8dc4e83c 100644
--- a/docs/tutorial/image_det.md
+++ b/docs/tutorial/image_det.md
@@ -67,7 +67,7 @@ If configuration finishes, the configuration file is generated in the `objectdet
 Train your model by running `blueoil train` with a model configuration.
 
 	$ docker run --rm \
-	    -e CUDA_VISIBLE_DEVICES=0 \
+	    --gpus '"device=0"' \
 	    -v $(pwd)/openimages_face:/home/blueoil/openimages_face \
 	    -v $(pwd)/config:/home/blueoil/config \
 	    -v $(pwd)/saved:/home/blueoil/saved \
@@ -75,7 +75,8 @@ Train your model by running `blueoil train` with a model configuration.
 	    blueoil train -c config/objectdetection.py
 
 Just like init, set the value of `{TAG}` to the value obtained by `docker images`.
-Change the value of `CUDA_VISIBLE_DEVICES` according to your environment.
+Change the value of `--gpus '"device=<gpu indices>"'` according to your environment. 
+For detail check the official [documentation](https://docs.docker.com/config/containers/resource_constraints/#gpu).
 
 When training has started, the training log and checkpoints are generated under `./saved/{MODEL_NAME}`.
 The value of `{MODEL_NAME}` will be `{Configuration file}_{TIMESTAMP}`.
diff --git a/docs/tutorial/image_keypoint_detection.md b/docs/tutorial/image_keypoint_detection.md
index 9b37b40fe..75364604d 100644
--- a/docs/tutorial/image_keypoint_detection.md
+++ b/docs/tutorial/image_keypoint_detection.md
@@ -61,7 +61,7 @@ If configuration finishes, the configuration file is generated in the `keypoint_
 Train your model by running `blueoil train` with a model configuration.
 
     $ docker run --rm \
-        -e CUDA_VISIBLE_DEVICES=0 \
+        --gpus '"device=0"' \
         -e DATA_DIR=/home/blueoil \
         -e OUTPUT_DIR=/home/blueoil/saved \
         -v /storage/dataset/MSCOCO_2017:/home/blueoil/MSCOCO_2017 \
@@ -71,7 +71,8 @@ Train your model by running `blueoil train` with a model configuration.
         blueoil train -c config/keypoint_detection_demo.py
 
 Just like init, set the value of `{TAG}` to the value obtained by `docker images`.
-Change the value of `CUDA_VISIBLE_DEVICES` according to your environment.
+Change the value of `--gpus '"device=<gpu indices>"'` according to your environment. 
+For detail check the official [documentation](https://docs.docker.com/config/containers/resource_constraints/#gpu).
 
 When training has started, the training log and checkpoints are generated under `./saved/{MODEL_NAME}`.
 The value of `{MODEL_NAME}` will be `{Configuration file}_{TIMESTAMP}`.
diff --git a/docs/tutorial/image_seg.md b/docs/tutorial/image_seg.md
index 8db3500c9..c31340026 100644
--- a/docs/tutorial/image_seg.md
+++ b/docs/tutorial/image_seg.md
@@ -97,7 +97,7 @@ If configuration finishes, the configuration file is generated in the `camvid.py
 Train your model by running `blueoil train` command with model configuration.
 
     $ docker run --rm \
-	    -e CUDA_VISIBLE_DEVICES=0 \
+	    --gpus '"device=0"' \
 	    -v $(pwd)/CamVid:/home/blueoil/CamVid \
 	    -v $(pwd)/config:/home/blueoil/config \
 	    -v $(pwd)/saved:/home/blueoil/saved \
@@ -105,7 +105,8 @@ Train your model by running `blueoil train` command with model configuration.
 	    blueoil train -c config/camvid.py
 
 Just like init, set the value of `{TAG}` to the value obtained by `docker images`.
-Change the value of `CUDA_VISIBLE_DEVICES` according to your environment.
+Change the value of `--gpus '"device=<gpu indices>"'` according to your environment. 
+For detail check the official [documentation](https://docs.docker.com/config/containers/resource_constraints/#gpu).
 
 When training has started, the training log and checkpoints are generated under `./saved/{MODEL_NAME}`.
 The value of `{MODEL_NAME}` will be `{Configuration file}_{TIMESTAMP}`.