Does TensorRT will support INT4 for Ampere architecture in the future ??
Related to this link it’s seems that INT4 can brings high performance improvements : https://developer.nvidia.com/blog/nvidia-ampere-architecture-in-depth/
Does TensorRT will support INT4 for Ampere architecture in the future ??
Related to this link it’s seems that INT4 can brings high performance improvements : https://developer.nvidia.com/blog/nvidia-ampere-architecture-in-depth/