Just curious why do we need qint8 when there is already int8? Is it because qint8 has a different and more efficient binary layout than that of the int8? Thanks!
- int8 is an integer type, it can be used for any operation which needs integers
- qint8 is a quantized tensor type which represents a compressed floating point tensor, it has an underlying int8 data layer, a scale, a zero_point and a qscheme
One could use torch.int8 as a component to build quantized int8 logic, that’s not how PyTorch does it today but we actually plan to converge towards this approach in the future.
1 Like