NGC | Catalog
Welcome Guest
CatalogContainersNeural Network Libraries

Neural Network Libraries

For pull tags and more information, please view on a desktop device.
Logo for Neural Network Libraries

Description

Neural Network Libraries is a deep learning framework that is intended to be used for research, development and production. We aim to have it running everywhere: desktop PCs, HPC clusters, embedded devices and production servers.

Publisher

Sony

Latest Tag

v1.24.0-mpi3.1.6

Modified

January 14, 2022

Compressed Size

4.71 GB

Multinode Support

Yes

Multi-Arch Support

No

What is Neural Network Libraries?

Neural Network Libraries is a deep learning framework that is intended to be used for research, development and production. We aim to have it running everywhere: desktop PCs, HPC clusters, embedded devices and production servers.

Neural Network Libraries is free for use

Neural Network Libraries is provided with a Apache license version 2.0. You could use, modify and re-publish Neural Network Libraries for free.

Quickstart Guide for Neural Network Libraries

  1. Pull NGC docker image
    $ docker pull ${image}
    
  2. Launch the container A typical way to start an interactive shell
    $ docker run --rm -ti -u $(id -u) -v /etc/passwd:/etc/passwd -v $(pwd):$(pwd) -v $HOME:$HOME -w $(pwd) --gpus all ${image} bash
    
    where
  • --rm Delete the container when exit
  • -ti Start container in interactive mode
  • -e Set envrionment variables
  • -u Username or UID
  • -v Bind mount a volume
  • -w Workding directory inside the container
  • --gpus GPU core numbers

Or to run specific task in the container and exit without interactive shell

$ docker run --rm -e -u $(id -u) -v /etc/passwd:/etc/passwd -v $(pwd):$(pwd) -v $HOME:$HOME -w $(pwd) --gpus all ${image} bash ./your_task.sh

where

  • --rm Delete the container when exit
  • -ti Start container in interactive mode
  • -e Set envrionment variables
  • -u Username or UID
  • -v Bind mount a volume
  • -w Workding directory inside the container
  • --gpus GPU core numbers
  • your_task.sh Your specified task
  1. Exit the container When you are done, close the container session by typing exit from container terminal.

What's included in the NGC image

Examples

The container contains in place working examples under /home/nnabla/examples, click example link for detailed README:

MINST Examples

  • mnist-collection

    Demostrate serveral deep learning algorithms on MNIST dataset, which is one of the most popular image classification datasets in the machine learning community. The MNIST dataset will be automatically downloaded when running any of the examples.

    • Classification task
      $ docker run --gpus all -w /home/nnabla/examples/mnist-collection ${image} python3 classification.py -c cudnn -i 1000
      
      $ docker run --gpus all -w /home/nnabla/examples/mnist-collection ${image} python3 classification_bnn.py -c cudnn -i 1000
      
    • Deep Convolutional GAN
      $ docker run --gpus all -w /home/nnabla/examples/mnist-collection ${image} python3 dcgan.py -c cudnn -i 1000
      
    • Semi-Supervised Learning of Classification
      $ docker run --gpus all -w /home/nnabla/examples/mnist-collection ${image} python3 vat.py -c cudnn -i 1000
      
    • Variational Auto-encoder
      $ docker run --gpus all -w /home/nnabla/examples/mnist-collection ${image} python3 vae.py -c cudnn -i 1000
      

Training Example using CIFAR-10 and CIFAR-100

  • cifar10-100-collection

    Demonstrate several deep learning algorithms on CIFAR-10 dataset and CIFAR-100 dataset, which are one of the most popular image classification datasets in the machine learning community. The datasets will be automatically downloaded when running any of the examples.

    • Classification task
      $ docker run --gpus all -w /home/nnabla/examples/cifar10-100-collection ${image} python3 classification.py -c cudnn -i 1000
      

Data Parallel Distributed Training Example using CIFAR-10

  • cifar10-100-collection/distributed:

    Demonstrate several deep learning algorithms on CIFAR-10 dataset, which is one of the most popular image classification datasets in the machine learning community. The CIFAR-10 dataset will be automatically downloaded when running any of the examples.

    • Multi-Device Multi-Process Training
      $ docker run --gpus all -w /home/nnabla/examples/cifar10-100-collection/distributed ${image} mpirun -n 4 python3 multi_device_multi_process_classification.py --context "cudnn" -b 64
      

Utils Frequently used

  • utils

    • Handly utils manipulating h5 and nnp

Additional Information

Github repositories

Known Issues

None

Technical Support

See official website community

Release Roadmap

Major components releases in the container, find their release dates and changelogs through the links: