Abstract
XTC changes the high-probability part of the next-token distribution. When multiple tokens pass a plausibility threshold, it can remove the dominant choices before sampling, providing a way to control diversity alongside temperature and repetition penalties.
Type
Publication
arXiv preprint
Ravid Shwartz Ziv
AI Researcher
AI researcher at Meta MSL with a background in information theory and computational neuroscience, working on world models, memory, and compression. Former Assistant Professor and Faculty Fellow at NYU. My industry experience spans Wand AI, Intel, Google AI, and Wikipedia.