2sang / oneshot-tfgpu

A little more Intuitive tool for manipulating a container for tensorflow, jupyter notebook

oneshot-tfgpu(WORK IN PROGRESS):

Yet another docker management tool for tensorflow-gpu & jupyter notebook

Getting Started

WIP

# Init new config settings with a few Q/A
tfgpu init

# Run last successful container
tfgpu run

# Start a container the image named 'c1'
tfgpu run c1

# Set image '1' with other option(s), changes will be saved locally at conf.yaml
# See 'tfgpu set --help' to see all available options
tfgpu set 1 --tag=latest-gpu-py3

# Apply 'c1' container changes to corresponding image
tfgpu commit c1

# List all pulled images
tfgpu ls

Todos & Features

• CircleCI, PyInquirer, python-fire
• Oneshot install only with a few configuration
• Stores multiple customized image instances -> restrict the number of countainer to one.
• Make tfgpu as direct shell command DONE
• argparse(or abseil) WIP
• Exception check for every scenario WIP
• Automate docker, nvidia-docker installation process WIP
• Image reset corrupted ones
• Docker commit, remove volume or supports other common commands.
• Docs for common use cases like ssh port forwarding, etc.