How we pick the patch size is kind of like inspired by how people do the tokenization but we don't want the tokenization at all so in our paper. For text we try to say don't stay too much away from bp token so right now bp token normally have one bp token corresponds to do around about four bytes and for image we kind of like pick patch of eight by eight or 16 by 16 that's what people do with the emitoconizer. Overall overall we find a wide range of patch size actually works quite well for our task so that's something definitely needs more analysis.

Get the Snipd
podcast app

Unlock the knowledge in podcasts with the podcast player of the future.
App store bannerPlay store banner

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode

Save any
moment

Hear something you like? Tap your headphones to save it with AI-generated key takeaways

Share
& Export

Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more

AI-powered
podcast player

Listen to all your favourite podcasts with AI-powered features

Discover
highlights

Listen to the best highlights from the podcasts you love and dive into the full episode