use cuda 10

Author: Will Feng

.circleci/config.yml

@@ -162,14 +162,14 @@ pytorch_tutorial_build_defaults: &pytorch_tutorial_build_defaults
 
 pytorch_tutorial_build_worker_defaults: &pytorch_tutorial_build_worker_defaults
   environment:
-    DOCKER_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-xenial-cuda9-cudnn7-py3"
-    CUDA_VERSION: "9"
+    DOCKER_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-xenial-cuda10-cudnn7-py3-gcc7"
+    CUDA_VERSION: "10"
   resource_class: gpu.medium
   <<: *pytorch_tutorial_build_defaults
 
 pytorch_tutorial_build_manager_defaults: &pytorch_tutorial_build_manager_defaults
   environment:
-    DOCKER_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-xenial-cuda9-cudnn7-py3"
+    DOCKER_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-xenial-cuda10-cudnn7-py3-gcc7"
   resource_class: medium
   <<: *pytorch_tutorial_build_defaults
 

.jenkins/build.sh

@@ -27,8 +27,8 @@ pip install sphinx-gallery tqdm matplotlib ipython pillow==4.1.1
 # Install torchvision from conda
 pip uninstall -y torchvision || true
 pip uninstall -y torchvision || true
-# GPU, requires CUDA version 9.0
-conda install https://anaconda.org/pytorch/torchvision/0.3.0/download/linux-64/torchvision-0.3.0-py36_cu9.0.176_1.tar.bz2 --no-deps
+# GPU, requires CUDA version 10.0
+conda install -c pytorch torchvision --no-deps
 
 # Install torchaudio from source
 git clone https://github.com/pytorch/audio --quiet
@@ -45,8 +45,8 @@ if [[ $(pip show torch) ]]; then
 fi
 
 # Install a nightly build of pytorch
-# GPU, requires CUDA version 9.0
-pip install cython torch_nightly -f https://download.pytorch.org/whl/nightly/cu90/torch_nightly.html
+# GPU, requires CUDA version 10.0
+pip install cython torch_nightly -f https://download.pytorch.org/whl/nightly/cu100/torch_nightly.html
 
 # Decide whether to parallelize tutorial builds, based on $JOB_BASE_NAME
 export NUM_WORKERS=20