Meta challenges transformer architecture with Megalodon LLM

Credit: VentureBeat using Midjourney


Megalodon also uses “chunk-wise attention,” which divides the input sequence into fixed-size blocks to reduce the complexity of the model from quadratic to linear.Read More