Deep-blueprint

Framework combining PyTorch Lightning and Hydra to make it easy to train a network.

You can use the code as-is, by adding your code to the different folders, but it is recommended to use it as a library, which is explained in the usage section.

Use as a library

Installing

First, you can install the code using pip :

pip install -e git+https://github.com/ispgroupucl/deep-blueprint.git@main#egg=deep_blueprint

WARNING : don't install this project in your global environment. First create a virtual environment using conda/virtualenv/venv/poetry/pyenv

NOTE : if you need to install a specific version of pytorch, install it before installing this project. The default pytorch version installed in this project doesn't work with the latest Nvidia gpu's (GeForce 3090, A100)

If you're using poetry, you can add the following line to pypoetry.toml :

deep_blueprint = { git = "https://github.com/ansible/ansible.git", develop = true}

Or, for pipenv, in the Pipfile file :

# torch = { version = "==1.11.0+cu113", index = "pytorch" } # If you need to use CUDA v11.3
# torchvision = { version = "==0.12.0+cu113", index = "pytorch" } # If you need to use CUDA v11.3
deep_blueprint = { editable = true, git = "https://github.com/ispgroupucl/deep-blueprint.git" }

Usage

TODO

Use directly

Installation

Using poetry :

poetry install

Running

To run a toy example of training on MNIST :

deep_blueprint gpus=1