✨ Test TensorFlow 2

Dockerfile

@@ -1,11 +0,0 @@
-FROM nvidia/cuda:10.0-cudnn7-runtime-ubuntu18.04
-RUN apt-get update &&\
-    apt-get install -y python3 python3-pip &&\
-    apt-get clean && rm -rf /var/lib/apt/lists/*
-COPY requirements.txt /tmp
-RUN pip3 install --no-cache-dir --upgrade pip && \
-    pip3 install --no-cache-dir -r /tmp/requirements.txt
-
-COPY test-nvidia /usr/bin
-
-CMD ["/usr/bin/test-nvidia"]

README.md

@@ -0,0 +1,7 @@
+Test Nvidia environment in relation to TensorFlow
+=================================================
+
+* `./run` tests the native system. One of tf1 or tf2 is expected to have no
+  GPU available due to CUDA library incompatibility
+* `./run-docker` tests Docker support. Both TensorFlow versions should work
+  as we're using a base image compatible to the respective version.

assets/Dockerfile

@@ -0,0 +1,19 @@
+ARG BASE_IMAGE
+
+FROM $BASE_IMAGE
+
+ARG tf
+
+RUN apt-get update && \
+    apt-get install -y python3 python3-distutils curl &&\
+    apt-get clean && rm -rf /var/lib/apt/lists/* && \
+    \
+    curl -sSL https://bootstrap.pypa.io/get-pip.py -o get-pip.py && \
+    python3 get-pip.py && \
+    rm -f get-pip.py
+COPY assets/requirements-$tf.txt /tmp
+RUN pip install --no-cache-dir -r /tmp/requirements-$tf.txt
+
+COPY test-nvidia /usr/bin
+
+CMD ["/usr/bin/test-nvidia"]

assets/requirements-tf2.txt

@@ -0,0 +1 @@
+tensorflow == 2.*

run

@@ -1,12 +1,16 @@
 #!/bin/sh
 
-vdir=`mktemp -d /tmp/test-nvidia.XXXXXX`
+for tf in tf1 tf2; do
+  echo "== $tf"
+  vdir=`mktemp -d /tmp/test-nvidia.XXXXXX`
 
-virtualenv -p /usr/bin/python3 $vdir
-. $vdir/bin/activate
+  # Need Python 3.7 here as TF1 does not support 3.8
+  virtualenv -q -p /usr/bin/python3.7 $vdir >/dev/null
+  . $vdir/bin/activate
 
-pip3 install -r requirements.txt
-python3 test-nvidia
+  pip install -q -r assets/requirements-$tf.txt
+  python3 test-nvidia
 
-deactivate
-rm --preserve-root -rf $vdir
+  deactivate
+  rm --preserve-root -rf $vdir
+done

run-docker

@@ -1,3 +1,14 @@
 #!/bin/sh
-docker build -t test-nvidia `dirname $0`
-docker run --gpus all -it --rm test-nvidia
+set -e
+
+for tf in tf1 tf2; do
+  echo "== $tf"
+  case "$tf" in
+    tf1) BASE_IMAGE=nvidia/cuda:10.0-cudnn7-runtime-ubuntu18.04;;
+    tf2) BASE_IMAGE=nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04;;
+  esac
+  work_dir=`dirname $0`
+  image_id=`docker build -q --build-arg tf=$tf --build-arg BASE_IMAGE=$BASE_IMAGE -f assets/Dockerfile $work_dir`
+  docker run --gpus all -it --rm $image_id
+  docker rmi $image_id >/dev/null || true
+done

test-nvidia

@@ -1,13 +1,21 @@
 #!/usr/bin/python3
 import os
-import tensorflow as tf
 
 
-os.system('nvidia-smi')
+os.system('nvidia-smi -L')
 
 
-hello = tf.constant('Hello, TensorFlow!')
-sess = tf.compat.v1.Session()
-print(sess.run(hello))
-print('GPU available:', tf.test.is_gpu_available(cuda_only=True))
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '1'  # '1' means >= WARN
+import tensorflow as tf
+print('TensorFlow', tf.__version__)
+with tf.compat.v1.Session() as sess:
+    hello = tf.constant('Hello, TensorFlow!')
+    result = sess.run(hello)
+    #print(result)
 
+if hasattr(tf.config, 'list_physical_devices'):
+    # TensorFlow 2
+    is_gpu_available = len(tf.config.list_physical_devices('GPU')) > 0
+    print('GPU available:', is_gpu_available)
+else:
+    print('GPU available:', tf.test.is_gpu_available(cuda_only=True))