Next-Concept Modeling

Text written coarse to fine, in your browser. Each level fills in a longer draft in parallel, one pass per level.

Paper Blog Code Models (Hazy Research) Larger models on GPU

Settings
The output appears here, coarse to fine.
Level1
Models and method by Hazy Research (MIT). Weights converted for the browser: ncm-webgpu-weights. @realmrfakename