nn-quantization-pytorch

nn-quantization-pytorch copied to clipboard

``` if self.bit_weights is not None: self.weight_quantization_default = quantization_mapping[self.qtype](self, self.weight, self.bit_weights, symmetric=True, uint=True, kwargs=kwargs) ``` I dont understand for a zero-centered weights, why quantize it as uint?

qiulinzhang

I'm trying to generate scales for resnet18 with 5 bits for weights and 7 bits for activation. When bn_folding is enabled there was a considerable drop in the accuracy. I...

NilakshanKunananthaseelan

Hi, It seems that your code of LAPQ in [here](https://github.com/ynahshan/nn-quantization-pytorch/blob/9a6488e9a103a5685afe799c3537f5665789c7ea/quantization/posttraining/cnn_classifier.py#L82-L90) use test dataset to do the calibration, does that indicate test dataset is used to obtain the optimal quantization step...

yhhhli

FileNotFoundError: Couldn't find any class folder in `/home/cvds_lab/datasets/ILSVRC2012/val`

2

I have downloaded the validation dataset `[Validation images (all tasks)](https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar). 6.3GB. MD5: 29b22e2961454d5413ddabcf34fc5622` from [https://image-net.org/challenges/LSVRC/2012/2012-downloads.php](https://image-net.org/challenges/LSVRC/2012/2012-downloads.php) I ran `python lapq/lapq_v2.py -a resnet18 -b 512 --dataset imagenet --pretrained --custom_resnet --min_method Powell -maxi...

PalgunaGopireddy