A python/pytorch utility library
- v0.0.3.2 available.
- v0.0.2 available. Added new datasets and quantization!
- v0.0.1 available. Feedbacks are highly welcomed!
conda install scopex/label/ScopeX::torchonn-pyutilspip install torchonn-pyutils --no-build-isolationor install from cloned codes from github if you would like to modify the code (recommend)
git clone https://github.com/JeremieMelo/pyutility.git
cd pyutility
./setup.shTo remove the package:
pip uninstall torchonn_pyutilsimport pyutils- Support pytorch training utility and datasets.
- Support lr_scheduler
- Support trainer
- Python >= 3.10
- PyTorch >= 2.0
- Tensorflow >= 2.5.0
- Others are listed in requirements.txt
| File | Description |
|---|---|
| datasets/ | Defines different datasets and builder |
| loss/ | Defines different loss functions/criterions |
| optimizer/ | Defines different optimizers |
| lr_scheduler/ | Defines different learning rate schedulers |
| quant/ | Defines different weight/activation quantizers |
| activation.py | Activation functions |
| compute.py | functions related to computing |
| config.py | Hierarchical yaml configuration file parser |
| distribution_sampler.py | Sample from customized distributions |
| general.py | Common helper functions |
| initializer.py | Initialization methods for PyTorch Parameters |
| loss.py | Loss functions for PyTorch model training |
| quantize.py | Quantization functions |
| torch_train.py | Helper functions for torch training |
| typing.py | Defines common types |
Jiaqi Gu ([email protected])