Current Jetson Nano image is based on Ubuntu distro, This project will try to deploy a opensuse version. Furthermore, I will take a closer look on deep learning framework, and learn how they use hardware accelerator.

First, boot up Jeston nano with Ubuntu, and deploy Tensorflow(Keras), Pytorch(Caffee2), MXNet, the most popular DL framework today, on it. Understand how those frameworks take advantage of hardware accelerator.

Second, build a new image with our kernel and rootfs.

Last, try to install DL frameworks from our ARM64 repo, checking current status.

Reference:

https://elinux.org/Jetson/Nano/Upstream

Looking for hackers with the skills:

Nothing? Add some keywords!

This project is part of:

Hack Week 18

Activity

  • almost 2 years ago: afesta liked this project.
  • almost 2 years ago: lyan originated this project.

  • Comments

    • lyan
      almost 2 years ago by lyan | Reply

      First part is almost done. Board is booted, and DL framworks are installed. Just could not find so much information about how its GPU work with arm cpu. There is a PCIe controller inside, maybe it is used pcie or just AXI bus, really wish its driver is open source, add-emoji

    • lyan
      almost 2 years ago by lyan | Reply

      Had an upgrade issue with Ubuntu, "files list file for package 'libacl1:arm64' is missing final newline", the solution is removing all "libacl1:arm64*" from "/var/lib/dpkg/info", there will be more same errors, just repeat it.

    • lyan
      almost 2 years ago by lyan | Reply

      Did a investigation on DL framworks, just some basic stuff, but it's good to me since I am more interested in how they work with hardware accelerator or how to improve. Most of them could work with Nvidia GPU by CUDA, and a few could work with AMD GPU and FPGA with OpenCL, and few of them could be supported directly by ASICs(TPU,NPU).

      =================== Tensorflow(google), most popular today, current version 2.0,

      Pytorch(FB), caffe2 is merged in pytorch now.

      Mxnet(Amazon, Nvidia)

      There are some others:

      Theano: stop develop since 2017

      Keras, user-friendly API for tensorflow and theano

      CNTK(MicroSoft), Cognitive Toolkit

      FastAI, A library based on Pytorch

      Reference: https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html https://www.tensorflow.org/learn https://pytorch.org/get-started/locally/ https://mxnet.apache.org/versions/master/architecture/index.html

    • lyan
      almost 2 years ago by lyan | Reply

      a pretty useful link:

      https://developer.nvidia.com/embedded/linux-tegra

    Similar Projects

    This project is one of its kind!