https://github.com/inferno-pytorch/speedrun-springboard

Boilerplate repository showing how to speedrun from idea to a reproducible experiment setup.
https://github.com/inferno-pytorch/speedrun-springboard

Last synced: 3 months ago
JSON representation

Boilerplate repository showing how to speedrun from idea to a reproducible experiment setup.

• Host: GitHub
• URL: https://github.com/inferno-pytorch/speedrun-springboard
• Owner: inferno-pytorch
• Created: 2019-05-27T20:44:28.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-05-28T14:22:55.000Z (over 5 years ago)
• Last Synced: 2024-04-24T16:43:13.053Z (7 months ago)
• Language: Python
• Size: 4.88 KB
• Stars: 1
• Watchers: 5
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# speedrun-springboard

Boilerplate repository showing how to speedrun from idea to a reproducible experiment setup with Pytorch.

This repository contains code to train a ResNet on Cifar-10. To get started: [install speedrun](https://github.com/inferno-pytorch/speedrun), [fork this repository](https://stackoverflow.com/questions/10065526/github-how-to-make-a-fork-of-public-repository-private) and [hack away](https://pytorch.org/)! Make sure to also install [tensorboardX](https://github.com/lanpa/tensorboardX) while you're at it. 

## Run Example

To run the example code on CPU, do the following. 

```bash

cd speedrun-springboard

mkdir experiments

python train.py experiments/FIRST-0 --inherit templates/BASE-X

```

This will make a new experiment directory `experiments/FIRST-0` (see [speedrun](https://github.com/inferno-pytorch/speedrun) for more on that). 

Now to train on GPU with a different learning rate, you could do the following: 

```bash

python train.py experiments/FIRST-1 --inherit experiments/FIRST-0 --config.device 'cuda:0' --config.optimizer.kwargs.lr 0.01

```

This will make another experiment directory `experiments/FIRST-1`. 

### Happy Hacking!