boilr 0.6.1

Basic framework for training models with PyTorch

boiler-pytorch

Basic framework for training stuff in PyTorch. It's quite tailored to projects I've been working on lately, so it's meant for personal use. Its sole purpose is to do away with boilerplate code, and having it here makes it easier to share it across projects.

Install

pip install boilr

Usage example/template

There's a usage example that can be useful as template. It's a basic VAE for MNIST quickly hacked together. The example files/folders are:

• example.py
• models/
• experiments/

Install requirements and run the example:

pip install -r requirements.txt
CUDA_VISIBLE_DEVICES=0 python example.py

Tested with:

• python 3.7.6
• numpy 1.18.1
• matplotlib 3.1.3
• torch 1.4.0
• torchvision 0.5.0
• tensorboard 2.1.0 (it also works without, but it won't save tensorboard logs)
• pillow 7.0
• tqdm 4.41.1