Stable, Fast and Accurate: Kernelized Attention with Relative Positional Encoding

Open in new window