fastai

The fastai deep learning library.

Copyright 2017 onwards, fast.ai, Inc. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. A copy of the License is provided in the LICENSE file in this repository.

Current Status

This is a ground-up rewrite of fastai. Everything should work, although docs are still in progress. If you're interested in contributing, join the discussion at: http://forums.fast.ai/c/fastai-dev.

Install

To use the notebooks or the beta version of the fastai modules you will need:

to use python 3.7 or python 3.6 with dataclasses: pip install dataclasses
to use the pytorch-nightly conda package, or the master branch of pytorch master
to install fastprogress: pip install fastprogress

PyPI Install

First install the nightly pytorch build, e.g. for CUDA 9.2:

pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu92/torch_nightly.html

If you have a different CUDA version find the right build here. Choose Preview/Linux/Pip/python3.6 and Your CUDA version and it will give you the correct install instruction.

Now you can install fastai. Note, that this is a beta test version at the moment, please report any issues:

 pip install --index-url https://test.pypi.org/simple/ --extra-index-url  https://pypi.org/simple/ fastai==1.0.0b3

Conda Install

First, find out also which CUDA version you need to install [here] (https://pytorch.org/get-started/locally/).

For example for CUDA9.2 it will be:

conda install cuda92 -c pytorch

You can also search the available cuda packages with:

conda search "cuda*" -c pytorch

If your setup doesn't have CUDA-support you will most likely have to wait till pytorch.org releases 1.0.0 packages for pytorch and torchvision.

Then, install the nightly pytorch build:

conda install -c pytorch pytorch-nightly

Once pytorch-nightly and the correct cudaXX package are installed then comes torchvision:

conda install -c fastai/label/test torchvision=0.2.1=pyhe7f20fa_0

We had to build a special version of torchvision which depends on pytorch-nightly.

Now you can install fastai. Note, that this is a beta test version at the moment, please report any issues.

Currently only linux-64/python3.6 conda build is available:

conda install -c fastai/label/test fastai

For other setups/platforms use pip install at the moment (see above).

Developer Install

First, follow the instructions above for either PyPi or Conda to install cudaXX, pytorch-nightly and torchvision, but don't install the pre-packaged fastai. Instead in the last step do:

git clone https://github.com/fastai/fastai_pytorch
cd fastai_pytorch
pip install -e .
tools/run-after-git-clone

Please refer to CONTRIBUTING.md and the developers guide for more details.

elyase / fastai_pytorch