https://github.com/Erlemar/pytorch_tempest

My repo for training neural nets using pytorch-lightning and hydra
https://github.com/Erlemar/pytorch_tempest

deep-learning hacktoberfest hydra pytorch-lightning training-pipeline

Last synced: about 2 months ago
JSON representation

My repo for training neural nets using pytorch-lightning and hydra

• Host: GitHub
• URL: https://github.com/Erlemar/pytorch_tempest
• Owner: Erlemar
• License: mit
• Created: 2020-08-30T10:46:59.000Z (about 4 years ago)
• Default Branch: master
• Last Pushed: 2023-10-22T13:46:01.000Z (11 months ago)
• Last Synced: 2024-04-17T01:18:13.875Z (5 months ago)
• Topics: deep-learning, hacktoberfest, hydra, pytorch-lightning, training-pipeline
• Language: Python
• Homepage: https://pytorch-tempest.readthedocs.io/en/latest/
• Size: 114 MB
• Stars: 201
• Watchers: 8
• Forks: 21
• Open Issues: 5
• Metadata Files:
  • Readme: README.md
  • Contributing: contributing.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# tempest

[![DeepSource](https://static.deepsource.io/deepsource-badge-light-mini.svg)](https://deepsource.io/gh/Erlemar/pytorch_tempest/?ref=repository-badge)

This repository has my pipeline for training neural nets.

Main frameworks used:

* [hydra](https://github.com/facebookresearch/hydra)

* [pytorch-lightning](https://github.com/PyTorchLightning/pytorch-lightning)

The main ideas of the pipeline:

* all parameters and modules are defined in configs;

* prepare configs beforehand for different optimizers/schedulers and so on, so it is easy to switch between them;

* have templates for different deep learning tasks. Currently, image classification and named entity recognition are supported;

Examples of running the pipeline:

This will run training on MNIST (data will be downloaded):

```shell

>>> python train.py --config-name mnist_config model.encoder.params.to_one_channel=True

```

Running on MPS (M1 macbook)

```shell

python train.py --config-name mnist_config model.encoder.params.to_one_channel=True trainer.accelerator=mps +trainer.devices=1 optimizer=adan training.lr=0.001

```

The default run:

```shell

>>> python train.py

```

The default version of the pipeline is run on imagenette dataset. To do it, download the data from this repository:

https://github.com/fastai/imagenette

unzip it and define the path to it in conf/datamodule/image_classification.yaml path