search

dongyp13 / Stochastic-Quantization

Training Low-bits DNNs with Stochastic Quantization
73 stars 18 forks source link

Pretrained models #1

Closed gudovskiy closed 7 years ago

gudovskiy commented 7 years ago

@dongyp13 Hi, thanks for sharing code. Could you release your pretrained models (specifically ImageNet AlexNet/ResNet SQ-TWN)? I try to reproduce your results but accuracy is very low even at first iterations...

dongyp13 commented 7 years ago

Hi, it's true that the accuracy is very low at first iterations (even at the first three quantization stages), but it will finally converge at the last stage with r = 100%. I will release the pretrain models after a few days.

dongyp13 commented 7 years ago

Hi @gudovskiy, our models are trained on another version of codes, which are not compatible with those on GitHub. We remove some functions and APIs in the GitHub version to make them clearer. You can try to train the model yourself from scratch. If you have some problems of training, you can email me to ask for the original version of codes and models.