How to adjust the VTA inp&wgt dtype from int8 to int16


#1

I need a more accurate VTA accelerator, to do some compution with 16bit integer input and weight. Could you please help me to point how to modify the VTA FPGA source? And are there some thing else need to change?


#2

somebody kindly help me?


#3

@cyberniklee one solution is to modify the config file: https://docs.tvm.ai/vta/dev/config.html before re-building the VTA design. One issue however is that some paramterizations might break the compilation flow. It’s brittle right now due to the vast number of parameters that we could use in the config. If you run into an error, please post it and we can see how to work around it.