• XCompression
• • Environment
  • First Use
  • Included method
  • • Pruning
    • Quantization
  • Reference
  • License

XCompression

An easy use Neural Network Compression framework.

Environment

Python

Pytorch

yacs

First Use

chmod 744 -R ./
pip install -r requirements.txt
python setup.py develop --user

We use YAML format to get configuration. All configs are listed in xcom/core/configs.py. Config files are listed in configs/

Included method

Pruning

Type	Name	Paper
Sparsity	Level
Channels/Sparsity	L1Norm	PRUNING FILTERS FOR EFFICIENT CONVNETS
Channels	FPGM	Filter Pruning via Geometric Median for Deep Convolutional Neural Networks Acceleration
Channels	Slim	Learning Efficient Convolutional Networks through Network Slimming
Channels	APoZ	Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architectures
Channels	Taylor	Importance Estimation for Neural Network Pruning

Quantization

Type	Name	Paper
	Default
	QAT	Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference
	DoReFa	DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients
	BNN	Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1
	LSQ	Learned Step Size Quantization

Reference

This project is based on microsoft/nni and facebookresearch/pycls

License

The entire codebase is under MIT license