Unifying Attention Heads and Task Vectors via Hidden State Geometry in In-Context Learning

Devs

Unifying Attention Heads and Task Vectors via Hidden State Geometry in In-Context Learning | Read Paper on Bytez