brevitas
brevitas copied to clipboard
FlexML quantized AvgPool update
Switch over to a Vitis-AI like implementation and export flow for QuantAvgPool2d for FlexML graph quantization. Keep the conversion of AvgPool to dephwise Conv as an option but simplify the flow by having a single scalar weight in the conv layer and by directly doing a AvgPool->QuantConv without passing through float Conv.
@cmcgirr-amd
Just an note (don't know if you're aware), I tried using a newer torch version than 1.10.0 and the export fails due to the onnx export API having changed keyword arguments