tf_lightning
This is tf_lightning, a very simple-small light-weight wrapper over tensorflow2, aims to train tf2 models with much less boiler-plate code and in a very structurized fashion.
Supported Features
It's currently supporting simple features for training models. I will add more features in this project as and when I need them.
- Single GPU training with much less code
- Mixed precision based training
- Distributed Training over multiple GPU's in single machine
- Gradient Accumulation
- Wandb integration
- Tensorboard integration
Installation
This small package, I designed for my personal use mainly. But if you are interested in using it, feel free to raise an issue; I will make docs regarding using it.
# install tensorflow-2.3 first
pip install tensorflow==2.3
# Run this command to install tf_lightning
pip install git+https://github.com/VasudevGupta7/tf-lightning.git@masterContributions
Feel free to fork this repositary and contribute to this project. Make sure you make pull request only in magik branch.