Bfloat16 quantization in tensorrt

Does TensorRT currently support quantization to bfloat16? I don’t see such an option. If not will there be that option in the future? Thanks!

Hi,

It’s not supported right now.
Based on the below link, it is on our list:

Thanks.