Moondream AI co-founder Vikhyat K. argues high speculative decoding acceptance rates signal inefficient tokenization in base models

VI

vik@vikhyatkTECH

@charles_irl @alisawuffles already fixed tokenizers. i just wish more models would adopt it:( https://arxiv.org/abs/2503.13423 but totally agree with your post. companies willing to do post-training can see a lot of gains here