!! Though this project can run on any device, it will be very slow. Unfortunately, it is really only dependent on the capabilities of your device as to how fast it can possibly run, but for a little better performance, run it here: https://turbowarp.org/1298164654?fps=250&turbo This is a prototype of a language model built with the PWSC string similarity algorithm. It can continue any string or also generate off of no input. It does not run very efficiently, as this is essentially the highest end version I could produce. The dataset comes from the Simple English Wikipedia. I have looked through it and cleaned it, and am almost entirely sure it is safe, but please message me if you find any line that has content that should be censored. The temperature variable controls how likely the model is to continue text in a way not necessarily reflective of the text's content or in an unpredictable manner. Low but nonzero temperature is recommended to keep text variable.
