I successfully pulled a gpu enabled docker image with
nvidia-docker run --rm -p 10000:8888 -e JUPYTER_ENABLE_LAB=yes -v "$PWD":/workdir isbjornlabs/fastai-notebook-cuda9.2
This image is based on the fastai lesson plan and includes all the libraries.
I start a new notebook and try to follow along the steps in lesI changed the batch size with
bs=16
son 1
This line crashes the kernel at about 6% into the first epoch.
learn.fit_one_cycle(4)
I changed the batch size with
bs=16
but it still crashes.
My gpu is a titanx pascal so I dont know why it should crash. I can run the same notebook fine in a kaggle kernel.