RoPE and ALiBi can be seen as KDA variants with imaginary or real log decay

teortaxesTex · x · 2026-07-28

A short technical claim ties several position-encoding ideas together:

The post is really a compact theory note about how different sequence models encode position, and it points toward a unified view rather than isolated tricks.

Related event: Kimi K3 Architecture Drops RoPE for NoPE(3 posts)→

Original post →

More from Research

Research channel →