Find a catchy name

[AmitMY]

The current name is image-latent-transformer which we abbreviate as ILT.
This is not so cool, and a better name would be LIT (latent-image-transformer).

But really, what is latent here? it is words, not specifically the image. The input representation can be a byte sequence or an image, or an audio of a word etc. So maybe this is a word-latent-transformer to really contrast byte-latent-transformer so perhaps we should go with WLT?

[AmitMY]

Name proposal by Timothee Mickus from Helsinki:
VisTA: Visual Text Architecture
OR
VisTA: Visual Input Space Transformer Architecture

Our "tokens" are not tokens or patches, they are windows.

Notate any exponentiation as XP(..)

Google "Windows buzzwords" and sprinkle them in

[AmitMY]

There are at least 5 papers in 2025 with the title Vista.

I am leaning towards toning down the "image" side and going more towards "words".

Maybe 🌍WeLT: Word Latent Transformer since we operate on words, as images, as bytes, maybe as signs or audio

[agblaszczyk]

Looking at this image makes me want to play around with the words lens or frame (especially thinking of the 'windows' you mentioned above)... Just a thought.

[elliottd]

Not sure about the importance of which concepts are important. I tried out some ideas combining the following terms: bytes, images, pixels, latent, model, transformer.

BIT: Byte-Image Transformer (perhaps too close to BitNet?)
BILT: Byte-Image Latent Transformer (perhaps too close to BLT?)
BIGM: Byte-Image Generative Model
PBLT: Pixel & Byte Latent Transformer
PBLM: Pixel & Byte Language Model