You'll need to run this with a Docker capable host computer, installation support is provided on Ubuntu.
install-docker.sh -- this will get Docker up and running
Now you can get a container ready with:
docker build --tag keras-cpu . docker run -p 8888:8888 --volume $(pwd):/src keras-cpu gpu/install-nvidia-docker.sh -- this requires you have a NVIDIA graphics card as well as the current driver.
If all is well at this point, you will see an inventory of your graphics cards such as:
+-----------------------------------------------------------------------------+ | NVIDIA-SMI 390.77 Driver Version: 390.77 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | |===============================+======================+======================| | 0 TITAN V Off | 00000000:03:00.0 Off | N/A | | 37% 53C P0 38W / 250W | 0MiB / 12066MiB | 0% Default | +-------------------------------+----------------------+----------------------+ | 1 TITAN V Off | 00000000:04:00.0 Off | N/A | | 33% 48C P0 37W / 250W | 0MiB / 12066MiB | 0% Default | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: GPU Memory | | GPU PID Type Process name Usage | |=============================================================================| | No running processes found | +-----------------------------------------------------------------------------+ Now you can get a container ready with:
docker build --tag keras-gpu ./gpu nvidia-docker run -p 8888:8888 --volume $(pwd):/src keras-gpu