search

zhutmost / lsq-net

Unofficial implementation of LSQ-Net, a neural network quantization framework
MIT License
276 stars 40 forks source link

unnecessary if-else or mistake? #16

Closed ice-tong closed 2 years ago

ice-tong commented 2 years ago

https://github.com/zhutmost/lsq-net/blob/2c24a96be06d044fa4c7d651727f4574b8d88c86/quan/quantizer/lsq.py#L48-L52

zhutmost commented 2 years ago

because the per_channel quantization option is incomplete.

A newer version is merged to https://github.com/zhutmost/neuralzip, and I removed this feature.

zhutmost commented 2 years ago

If you want to quantize per output channel, you need to adjust the hyper parameters. Or the optimizer may not converge.

ice-tong commented 2 years ago

Thanks for your reply.