If the two input nodes of add
op share the same quantization params(scale and zero point), the calculation of quantized_add
will be simpler. Can multiple nodes share the same observer in FX Graph Mode Quantization?
How to let the two input nodes of `add` op share the same quantization params(scale and zero point)?
we don’t support this in fx, but the new quant flow will support this: How to Write a Quantizer for PyTorch 2 Export Quantization — PyTorch Tutorials 2.1.0+cu121 documentation
Thanks for your reply.