Upgrade the Tensorflow version to address some security issues (#1870)

* upgrade the tensorflow version to address some security issues

* fix enas example codes

* upgrade tensorflow to v2.9.1 and tensorflow-aarch64 to v2.9.0

* install protobuf (>= 3.9.2, < 3.20) for tensorflow-aarch64

Author: tenzen-y

cmd/metricscollector/v1beta1/tfevent-metricscollector/Dockerfile

@@ -2,19 +2,11 @@ FROM python:3.9
 
 ENV TARGET_DIR /opt/katib
 ENV METRICS_COLLECTOR_DIR cmd/metricscollector/v1beta1/tfevent-metricscollector
-# tensorflow community build for aarch64
-# https://github.com/tensorflow/build#tensorflow-builds
-ENV PIP_EXTRA_INDEX_URL https://snapshots.linaro.org/ldcg/python-cache/
 
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${METRICS_COLLECTOR_DIR}/ ${TARGET_DIR}/${METRICS_COLLECTOR_DIR}/
 WORKDIR  ${TARGET_DIR}/${METRICS_COLLECTOR_DIR}
 
-RUN if [ "$(uname -m)" = "aarch64" ]; then \
-    pip install tensorflow-aarch64==2.7.0; \
-    else \
-    pip install tensorflow==2.7.0; \
-    fi;
 RUN pip install --no-cache-dir -r requirements.txt
 
 RUN chgrp -R 0 ${TARGET_DIR} \

cmd/metricscollector/v1beta1/tfevent-metricscollector/requirements.txt

@@ -2,3 +2,9 @@ psutil==5.8.0
 rfc3339>=6.2
 grpcio==1.41.1
 googleapis-common-protos==1.6.0
+# TODO (tenzen-y): We need to delete the line to install protobuf after tensorflow-aarch64 v2.9.1, or higher has been released.
+# To avoid the `If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.` error,
+# we must restrict the protobuf version.
+protobuf >= 3.9.2, < 3.20; platform_machine=="aarch64"
+tensorflow==2.9.1; platform_machine=="x86_64"
+tensorflow-aarch64==2.9.0; platform_machine=="aarch64"

cmd/suggestion/chocolate/v1beta1/Dockerfile

@@ -22,9 +22,6 @@ RUN if [ "$(uname -m)" = "ppc64le" ]; then \
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
-RUN if [ "$(uname -m)" = "aarch64" ]; then \
-    sed -i -e '$a git+https://github.com/fmder/ghalton@master' -e '/^ghalton/d' requirements.txt; \
-    fi;
 RUN pip install --no-cache-dir -r requirements.txt
 
 RUN chgrp -R 0 ${TARGET_DIR} \

cmd/suggestion/chocolate/v1beta1/requirements.txt

@@ -8,5 +8,6 @@ protobuf==3.19.1
 googleapis-common-protos==1.6.0
 SQLAlchemy==1.4.26
 git+https://github.com/AIworx-Labs/chocolate@master
-ghalton>=0.6.2
+ghalton>=0.6.2; platform_machine=="x86_64"
+git+https://github.com/fmder/ghalton@master; platform_machine=="aarch64"
 cython>=0.29.24

cmd/suggestion/nas/enas/v1beta1/Dockerfile

@@ -3,9 +3,6 @@ FROM python:3.9
 ENV TARGET_DIR /opt/katib
 ENV SUGGESTION_DIR cmd/suggestion/nas/enas/v1beta1
 ENV GRPC_HEALTH_PROBE_VERSION v0.4.11
-# tensorflow community build for aarch64
-# https://github.com/tensorflow/build#tensorflow-builds
-ENV PIP_EXTRA_INDEX_URL https://snapshots.linaro.org/ldcg/python-cache/
 
 RUN if [ "$(uname -m)" = "ppc64le" ] || [ "$(uname -m)" = "aarch64" ]; then \
     apt-get -y update && \
@@ -26,10 +23,6 @@ RUN if [ "$(uname -m)" = "ppc64le" ]; then \
 ADD ./pkg/ ${TARGET_DIR}/pkg/
 ADD ./${SUGGESTION_DIR}/ ${TARGET_DIR}/${SUGGESTION_DIR}/
 WORKDIR  ${TARGET_DIR}/${SUGGESTION_DIR}
-
-RUN if [ "$(uname -m)" = "aarch64" ]; then \
-    sed -i 's/tensorflow==/tensorflow-aarch64==/' requirements.txt; \
-    fi;
 RUN pip install --no-cache-dir -r requirements.txt
 
 RUN chgrp -R 0 ${TARGET_DIR} \

cmd/suggestion/nas/enas/v1beta1/requirements.txt

@@ -1,5 +1,9 @@
 grpcio==1.41.1
-protobuf==3.19.1
 googleapis-common-protos==1.6.0
-tensorflow==2.8.0
 cython>=0.29.24
+# TODO (tenzen-y): We need to delete the line to install protobuf after tensorflow-aarch64 v2.9.1, or higher has been released.
+# To avoid the `If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0.` error,
+# we must restrict the protobuf version.
+protobuf >= 3.9.2, < 3.20; platform_machine=="aarch64"
+tensorflow==2.9.1; platform_machine=="x86_64"
+tensorflow-aarch64==2.9.0; platform_machine=="aarch64"

examples/v1beta1/trial-images/enas-cnn-cifar10/Dockerfile.cpu

@@ -1,4 +1,4 @@
-FROM tensorflow/tensorflow:2.7.0
+FROM tensorflow/tensorflow:2.9.1
 
 ENV TARGET_DIR /opt/enas-cnn-cifar10
 

examples/v1beta1/trial-images/enas-cnn-cifar10/Dockerfile.gpu

@@ -1,4 +1,4 @@
-FROM tensorflow/tensorflow:2.7.0-gpu
+FROM tensorflow/tensorflow:2.9.1-gpu
 
 ENV TARGET_DIR /opt/enas-cnn-cifar10
 

examples/v1beta1/trial-images/enas-cnn-cifar10/RunTrial.py

@@ -16,8 +16,8 @@
 from keras.datasets import cifar10
 from ModelConstructor import ModelConstructor
 from tensorflow.keras.utils import to_categorical
-from tensorflow.python.keras.utils.multi_gpu_utils import multi_gpu_model
 from keras.preprocessing.image import ImageDataGenerator
+import tensorflow as tf
 import argparse
 
 if __name__ == "__main__":
@@ -50,16 +50,23 @@
 
     print("\n>>> Constructing Model...")
     constructor = ModelConstructor(arch, nn_config)
-    test_model = constructor.build_model()
-    print(">>> Model Constructed Successfully\n")
 
-    if num_gpus > 1:
-        test_model = multi_gpu_model(test_model, gpus=num_gpus)
+    num_physical_gpus = len(tf.config.experimental.list_physical_devices('GPU'))
+    if 1 <= num_gpus <= num_physical_gpus:
+        devices = ["/gpu:"+str(i) for i in range(num_physical_gpus)]
+    else:
+        num_physical_cpu = len(tf.config.experimental.list_physical_devices('CPU'))
+        devices = ["/cpu:"+str(j) for j in range(num_physical_cpu)]
+
+    strategy = tf.distribute.MirroredStrategy(devices)
+    with strategy.scope():
+        test_model = constructor.build_model()
+        test_model.summary()
+        test_model.compile(loss=keras.losses.categorical_crossentropy,
+                           optimizer=keras.optimizers.Adam(learning_rate=1e-3, decay=1e-4),
+                           metrics=['accuracy'])
 
-    test_model.summary()
-    test_model.compile(loss=keras.losses.categorical_crossentropy,
-                       optimizer=keras.optimizers.Adam(learning_rate=1e-3, decay=1e-4),
-                       metrics=['accuracy'])
+    print(">>> Model Constructed Successfully\n")
 
     (x_train, y_train), (x_test, y_test) = cifar10.load_data()
     x_train = x_train.astype('float32')

examples/v1beta1/trial-images/tf-mnist-with-summaries/Dockerfile

@@ -1,4 +1,4 @@
-FROM tensorflow/tensorflow:2.7.0
+FROM tensorflow/tensorflow:2.9.1
 
 ADD examples/v1beta1/trial-images/tf-mnist-with-summaries /opt/tf-mnist-with-summaries
 WORKDIR /opt/tf-mnist-with-summaries