Does NNI Supports symmetric quantization?

[twmht]

According to the TensorRT document (https://docs.nvidia.com/deeplearning/tensorrt/archives/tensorrt-700/tensorrt-developer-guide/index.html#set_tensor_mp_c), symmetric quantization is supported.

Therefore when using TensorRT, we would see the warnings

DynamicRange(min: -0.424211, max: 2.82148). Dynamic range should be symmetric for better accuracy)

But NNI only supports asymmetric quantization.

Any idea?

[linbinskn]

We have planned to support different dtypes and qschemes in this pr which includes symmetric quantization.

[twmht]

nice. hopefully this would be merge.

[chenbohua3]

working on it :)