Accessing accumulated quantized result

quantization
1

Hi,

The PyTorch quantized operations are great but they return the result after it has been quantized back to 8-bit integer. Is there a simple way of accessing the accumulated result at int32? Are there other libraries, such as Caffe or a different quantization backend, that have this API?

2

Re-quantization (conversion from int32 to int8) is performed in the quantized operator libraries - FBGEMM and QNNPACK.
Currently there is no easy way to access this value from pytorch API.