Next-token prediction in JavaScript — build fast language and diffusion models.

In-painting: Implement LSTM based pixel prediction to predict the missing pixel in an image

This repository is made as supplementary material for a tutorial. The tutorial shows how to use Recurrent Neural Nets as generative models.

This experimental work presents a different approach to increase the size and quality of an image by adding a blank pixel around each pixel in an image, enlarging the image, breaking it into parts, and generating these blank pixels by predicting them with models.