The first release of the package. You could fetch the package via `pip install tqt` or `conda install tqt` or [https://pypi.org/project/tqt/0.1.0/](https://pypi.org/project/tqt/0.1.0/).
Provide three thing:
1. some tqt-version functions of `torch.nn`
2. threshold initialization given by tqt paper
3. utils tools to make networks based on torch easily to be quantized
functions:
- (no quantization needed)MaxPool1d, MaxPool2d, MaxPool3d
- (no quantization needed)AvgPool1d, AvgPool2d, AvgPool3d
- ReLU, ReLU6
- BatchNorm2d
- Conv2d
- Linear
- (from addernet paper)Adder2d
threshold initialization
- init_network function to initialize the whole net
- 3sd init
- max init
- TensorRT's entropy calibration init
- hooks needed by the init
utils:
- make_net_quant_or_not, with exclusion list.