Installation of fastai2 on ubuntu lts machine without GPU locally

Hi,
I am planning to use a thin client(without GPU) with Ubuntu 18.04 for inference. I am new to Ubuntu and was earlier using Windows10 without GPU for inference. How to install fastai2 on the machine as well as import the fastbook?

Hi,
I always strongly recommend to use docker whenever possible. In case if Jupyter it works very, very well. You can get an image that already has fast.ai with all dependencies installed.

I assume you are not too familiar with docker and linux, hence I would recommend:

  1. Install docker on your Windows, and learn the basics commands: pull image start image, stop image, connect to the image with shell and explore what is inside. Learning docker on an OS that you already know will be easier, doing this on Linux will be overwhelming for sure.
    This can take a day or two. https://www.youtube.com/results?search_query=docker+tutorial+windows
    Get an image of a web app you already know and see how it behaves in container.

  2. Once you know how to download, start, stop and delete, get yourself the latest image with fastai: docker pull fastdotai/fastai:latest and start it locally

  3. If you’ve completed the above, then you can repeat it on linux.

Good luck !

2 Likes

Thanks for the suggestions. I feel this is the way to go ahead. However presently I am trying to solve a problem which is due to shift from fastai1 to fastai2. I have a grain analysis system which uses OpenCV and and two fastai1 models.The application invokes the python script and does not depend on Jupyter notebook. I have to replicate the system as well as train on new grain images. I plan to do the training on Colab or Gradient server with Fastai2. The trained model now needs to be deployed in a standalone machine which does only inferencing. If I can install Fastai2 on a ubuntu machine I can run the same python program with minimal changes.
Do you still recommend using docker ? Alternately please let me know how to deploy fastai2 on Ubuntu machine.
Regards

Hi,
For me, Docker is the biggest breakthrough in the non AI part of IT. It is here to grow and expand even more. I strongly recommend that you learn the basics and spend a day on it.

If you need to run Jupyter notebook installed on Ubuntu, with latest python, condas, all AI libraries including fastia, then the fastest and most stable way is to install docker and get the fastai image.

For you it will work like a small linux VM, with jupyter and all dependencies on it. You will get a localhost url to access jupyter.

1 Like