1 paper
Ingo Ziegler, Martin Krebs, Desmond Elliott
Language models encode text as subword tokens, raw bytes, or rendered pixels, but these encodings are usually compared under modeling constraints that expose different amounts of l…