Infrastructure to enable deployment of ML models to low-power resource-constrained embedded targets (including microcontrollers and digital signal processors).
From fully_connected_common.cc I see that filter weights must be symmetric, i.e. zero_point=0. How can I achieve this? Also, is it only possible by using quantization-aware training, or it can it be done during post-training quantization? Is there a reference or example I can refer to? Thanks
From
fully_connected_common.cc
I see that filter weights must be symmetric, i.e.zero_point=0
. How can I achieve this? Also, is it only possible by using quantization-aware training, or it can it be done during post-training quantization? Is there a reference or example I can refer to? Thanks