Mixed Precision Models

I have an ONNX model, in FP32 precision, which I would like to convert to a mixed precision FP16-INT8. Is that possible? If so, how would I use Nvidia tools to achieve that.


It’s possible.
You can convert it with our trtexec tool with the --best flag.
It will choose a faster algorithm for you directly.


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.