Reichman University · Data Science Institute Intellexus
← → reveal / navigate · O map · N notes · F fullscreen

Jump to a slide

Speaker notes
‹

Opening

Do We Really Need So Many Tokens? The Performance-Efficiency Trade-off in Tibetan and Sanskrit

Orr Meir Zwebner, Kai Golan Hashiloni, Guy Bilitski, Oded Hellman, Nir Marciano, Erez Yousfan, Shai Fine, Kfir Bar

Methodology scheme: Models, Languages, vocabulary expansion sizes, number of CPT samples, Train (embedding initialization, stage 1, stage 2), Evaluation, and the evaluation outputs.
or click anywhere · press → or Spaceor swipe ←
Best viewed in landscape — rotate your phone