[cherry-pick] Add the pytorch-mnist with GPU support container image (#…

…1917)
kubeflow · Jul 17, 2022 · 12a4896 · 12a4896
1 parent 8dcc7d3
commit 12a4896
Show file tree

Hide file tree

Showing 15 changed files with 61 additions and 25 deletions.
diff --git a/.github/workflows/publish-trial-images.yaml b/.github/workflows/publish-trial-images.yaml
@@ -31,8 +31,10 @@ jobs:
         include:
           - trial-name: mxnet-mnist
             dockerfile: examples/v1beta1/trial-images/mxnet-mnist/Dockerfile
-          - trial-name: pytorch-mnist
-            dockerfile: examples/v1beta1/trial-images/pytorch-mnist/Dockerfile
+          - trial-name: pytorch-mnist-cpu
+            dockerfile: examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.cpu
+          - trial-name: pytorch-mnist-gpu
+            dockerfile: examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.gpu
           - trial-name: tf-mnist-with-summaries
             dockerfile: examples/v1beta1/trial-images/tf-mnist-with-summaries/Dockerfile
           - trial-name: enas-cnn-cifar10-gpu

diff --git a/.github/workflows/pytorch-mnist-e2e-test.yaml b/.github/workflows/pytorch-mnist-e2e-test.yaml
@@ -24,7 +24,7 @@ jobs:
           experiments: ${{ matrix.experiments }}
           training-operator: true
           # Comma Delimited
-          trial-images: pytorch-mnist
+          trial-images: pytorch-mnist-cpu
 
     strategy:
       fail-fast: false

diff --git a/docs/images-location.md b/docs/images-location.md
@@ -273,13 +273,24 @@ The following table shows images for training containers which are used in the
     </tr>
     <tr align="center">
       <td>
-        <code>docker.io/kubeflowkatib/pytorch-mnist</code>
+        <code>docker.io/kubeflowkatib/pytorch-mnist-cpu</code>
       </td>
       <td>
-        PyTorch MNIST example with printing metrics to the file or StdOut
+        PyTorch MNIST example with printing metrics to the file or StdOut with CPU support
       </td>
       <td>
-        <a href="https://github.com/kubeflow/katib/blob/master/examples/v1beta1/trial-images/pytorch-mnist/Dockerfile">Dockerfile</a>
+        <a href="https://github.com/kubeflow/katib/blob/master/examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.cpu">Dockerfile</a>
+      </td>
+    </tr>
+    <tr align="center">
+      <td>
+        <code>docker.io/kubeflowkatib/pytorch-mnist-gpu</code>
+      </td>
+      <td>
+        PyTorch MNIST example with printing metrics to the file or StdOut with GPU support
+      </td>
+      <td>
+        <a href="https://github.com/kubeflow/katib/blob/master/examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.gpu">Dockerfile</a>
       </td>
     </tr>
     <tr align="center">

diff --git a/examples/v1beta1/early-stopping/median-stop-with-json-format.yaml b/examples/v1beta1/early-stopping/median-stop-with-json-format.yaml
@@ -62,7 +62,7 @@ spec:
           spec:
             containers:
               - name: training-container
-                image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                 command:
                   - "python3"
                   - "/opt/pytorch-mnist/mnist.py"

diff --git a/examples/v1beta1/kubeflow-training-operator/pytorchjob-mnist.yaml b/examples/v1beta1/kubeflow-training-operator/pytorchjob-mnist.yaml
@@ -46,7 +46,7 @@ spec:
               spec:
                 containers:
                   - name: pytorch
-                    image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                    image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                     command:
                       - "python3"
                       - "/opt/pytorch-mnist/mnist.py"
@@ -61,7 +61,7 @@ spec:
               spec:
                 containers:
                   - name: pytorch
-                    image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                    image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                     command:
                       - "python3"
                       - "/opt/pytorch-mnist/mnist.py"

diff --git a/examples/v1beta1/metrics-collector/custom-metrics-collector.yaml b/examples/v1beta1/metrics-collector/custom-metrics-collector.yaml
@@ -67,7 +67,7 @@ spec:
           spec:
             containers:
               - name: training-container
-                image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                 command:
                   - "python3"
                   - "/opt/pytorch-mnist/mnist.py"

diff --git a/examples/v1beta1/metrics-collector/file-metrics-collector-with-json-format.yaml b/examples/v1beta1/metrics-collector/file-metrics-collector-with-json-format.yaml
@@ -52,7 +52,7 @@ spec:
           spec:
             containers:
               - name: training-container
-                image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                 command:
                   - "python3"
                   - "/opt/pytorch-mnist/mnist.py"

diff --git a/examples/v1beta1/metrics-collector/file-metrics-collector.yaml b/examples/v1beta1/metrics-collector/file-metrics-collector.yaml
@@ -54,7 +54,7 @@ spec:
           spec:
             containers:
               - name: training-container
-                image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                 command:
                   - "python3"
                   - "/opt/pytorch-mnist/mnist.py"

diff --git a/...ta1/trial-images/pytorch-mnist/Dockerfile → ...trial-images/pytorch-mnist/Dockerfile.cpu b/...ta1/trial-images/pytorch-mnist/Dockerfile → ...trial-images/pytorch-mnist/Dockerfile.cpu
diff --git a/examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.gpu b/examples/v1beta1/trial-images/pytorch-mnist/Dockerfile.gpu
@@ -0,0 +1,15 @@
+FROM pytorch/pytorch:1.11.0-cuda11.3-cudnn8-runtime
+
+ADD examples/v1beta1/trial-images/pytorch-mnist /opt/pytorch-mnist
+WORKDIR /opt/pytorch-mnist
+
+# Add folder for the logs.
+RUN mkdir /katib
+RUN pip install --no-cache-dir -r requirements.txt
+
+RUN chgrp -R 0 /opt/pytorch-mnist \
+  && chmod -R g+rwX /opt/pytorch-mnist \
+  && chgrp -R 0 /katib \
+  && chmod -R g+rwX /katib
+
+ENTRYPOINT ["python3", "/opt/pytorch-mnist/mnist.py"]
diff --git a/manifests/v1beta1/components/controller/trial-templates.yaml b/manifests/v1beta1/components/controller/trial-templates.yaml
@@ -54,7 +54,7 @@ data:
             spec:
               containers:
                 - name: pytorch
-                  image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                  image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                   command:
                     - "python3"
                     - "/opt/pytorch-mnist/mnist.py"
@@ -68,7 +68,7 @@ data:
             spec:
               containers:
                 - name: pytorch
-                  image: docker.io/kubeflowkatib/pytorch-mnist:v0.14.0-rc.0
+                  image: docker.io/kubeflowkatib/pytorch-mnist-cpu:v0.14.0-rc.0
                   command:
                     - "python3"
                     - "/opt/pytorch-mnist/mnist.py"

diff --git a/scripts/v1beta1/build.sh b/scripts/v1beta1/build.sh
@@ -123,8 +123,11 @@ else
   echo -e "\nBuilding mxnet mnist training container example...\n"
   docker build --platform linux/amd64 -t "${REGISTRY}/mxnet-mnist:${TAG}" -f examples/${VERSION}/trial-images/mxnet-mnist/Dockerfile .
 
-  echo -e "\nBuilding PyTorch mnist training container example...\n"
-  docker build --platform linux/amd64 -t "${REGISTRY}/pytorch-mnist:${TAG}" -f examples/${VERSION}/trial-images/pytorch-mnist/Dockerfile .
+  echo -e "\nBuilding PyTorch mnist training container example with CPU support...\n"
+  docker build --platform linux/amd64 -t "${REGISTRY}/pytorch-mnist-cpu:${TAG}" -f examples/${VERSION}/trial-images/pytorch-mnist/Dockerfile.cpu .
+
+  echo -e "\nBuilding PyTorch mnist training container example with GPU support...\n"
+  docker build --platform linux/amd64 -t "${REGISTRY}/pytorch-mnist-gpu:${TAG}" -f examples/${VERSION}/trial-images/pytorch-mnist/Dockerfile.gpu .
 
   echo -e "\nBuilding Keras CIFAR-10 CNN training container example for ENAS with GPU support...\n"
   docker build --platform linux/amd64 -t "${REGISTRY}/enas-cnn-cifar10-gpu:${TAG}" -f examples/${VERSION}/trial-images/enas-cnn-cifar10/Dockerfile.gpu .

diff --git a/scripts/v1beta1/push.sh b/scripts/v1beta1/push.sh
@@ -98,8 +98,11 @@ docker push "${REGISTRY}/mxnet-mnist:${TAG}"
 echo -e "\nPushing Tensorflow with summaries mnist training container example...\n"
 docker push "${REGISTRY}/tf-mnist-with-summaries:${TAG}"
 
-echo -e "\nPushing PyTorch mnist training container example...\n"
-docker push "${REGISTRY}/pytorch-mnist:${TAG}"
+echo -e "\nPushing PyTorch mnist training container example with CPU support...\n"
+docker push "${REGISTRY}/pytorch-mnist-cpu:${TAG}"
+
+echo -e "\nPushing PyTorch mnist training container example with GPU support...\n"
+docker push "${REGISTRY}/pytorch-mnist-gpu:${TAG}"
 
 echo -e "\nPushing Keras CIFAR-10 CNN training container example for ENAS with GPU support...\n"
 docker push "${REGISTRY}/enas-cnn-cifar10-gpu:${TAG}"

diff --git a/scripts/v1beta1/update-images.sh b/scripts/v1beta1/update-images.sh
@@ -83,7 +83,8 @@ update_yaml_files "${CONFIG_PATH}" ":[^[:space:]].*\"" ":${TAG}\""
 
 # Postfixes for the each Trial image.
 MXNET_MNIST="mxnet-mnist"
-PYTORCH_MNIST="pytorch-mnist"
+PYTORCH_MNIST_CPU="pytorch-mnist-cpu"
+PYTORCH_MNIST_GPU="pytorch-mnist-gpu"
 TF_MNIST_WITH_SUMMARIES="tf-mnist-with-summaries"
 ENAS_GPU="enas-cnn-cifar10-gpu"
 ENAS_CPU="enas-cnn-cifar10-cpu"
@@ -93,7 +94,8 @@ SIMPLE_PBT="simple-pbt"
 
 echo -e "Update Katib Trial training container images\n"
 update_yaml_files "./" "${OLD_PREFIX}${MXNET_MNIST}:.*" "${NEW_PREFIX}${MXNET_MNIST}:${TAG}"
-update_yaml_files "./" "${OLD_PREFIX}${PYTORCH_MNIST}:.*" "${NEW_PREFIX}${PYTORCH_MNIST}:${TAG}"
+update_yaml_files "./" "${OLD_PREFIX}${PYTORCH_MNIST_CPU}:.*" "${NEW_PREFIX}${PYTORCH_MNIST_CPU}:${TAG}"
+update_yaml_files "./" "${OLD_PREFIX}${PYTORCH_MNIST_GPU}:.*" "${NEW_PREFIX}${PYTORCH_MNIST_GPU}:${TAG}"
 update_yaml_files "./" "${OLD_PREFIX}${TF_MNIST_WITH_SUMMARIES}:.*" "${NEW_PREFIX}${TF_MNIST_WITH_SUMMARIES}:${TAG}"
 update_yaml_files "./" "${OLD_PREFIX}${ENAS_GPU}:.*" "${NEW_PREFIX}${ENAS_GPU}:${TAG}"
 update_yaml_files "./" "${OLD_PREFIX}${ENAS_CPU}:.*" "${NEW_PREFIX}${ENAS_CPU}:${TAG}"

diff --git a/test/e2e/v1beta1/scripts/gh-actions/build-load.sh b/test/e2e/v1beta1/scripts/gh-actions/build-load.sh
@@ -30,7 +30,7 @@ REGISTRY="docker.io/kubeflowkatib"
 TAG="e2e-test"
 VERSION="v1beta1"
 CMD_PREFIX="cmd"
-SPECIFIED_DEVICE_TYPE_IMAGES=("enas-cnn-cifar10-cpu" "darts-cnn-cifar10-cpu")
+SPECIFIED_DEVICE_TYPE_IMAGES=("enas-cnn-cifar10-cpu" "darts-cnn-cifar10-cpu" "pytorch-mnist-cpu")
 
 IFS="," read -r -a TRIAL_IMAGE_ARRAY <<< "$TRIAL_IMAGES"
 IFS="," read -r -a EXPERIMENT_ARRAY <<< "$EXPERIMENTS"
@@ -51,7 +51,7 @@ _build_containers() {
   docker build --platform "$(uname -m)" -t "$REGISTRY/$CONTAINER_NAME:$TAG" -f "../../../../../$DOCKERFILE" ../../../../../
 }
 
-_load_kind_cluster() {
+_load_minikube_cluster() {
   CONTAINER_NAME=${1:-"katib-controller"}
 
   echo -e "\n\nLoading $CONTAINER_NAME image...\n\n"
@@ -99,7 +99,7 @@ run() {
     for s in "${suggestions[@]}"; do
       if [ "$s" == "$CONTAINER_NAME" ]; then
         _build_containers "$CONTAINER_NAME" "$DOCKERFILE"
-        _load_kind_cluster "$CONTAINER_NAME"
+        _load_minikube_cluster "$CONTAINER_NAME"
         break
       fi
     done
@@ -126,15 +126,15 @@ run() {
     for e in "${earlystoppings[@]}"; do
       if [ "$e" == "$CONTAINER_NAME" ]; then
         _build_containers "$CONTAINER_NAME" "$DOCKERFILE"
-        _load_kind_cluster "$CONTAINER_NAME"
+        _load_minikube_cluster "$CONTAINER_NAME"
         break
       fi
     done
 
   # Others
   else
     _build_containers "$CONTAINER_NAME" "$DOCKERFILE"
-    _load_kind_cluster "$CONTAINER_NAME"
+    _load_minikube_cluster "$CONTAINER_NAME"
   fi
 }