dss-projectfutura 0.2.4

Content Generation based on Caches

DSSv3g

This is the third version of the DSS text generator, but built in a more generic way to allow for any kind of dataset. Lots of new features have been implemented, helping DSSv3g become actually useful.

DSSv3g is still built using the caches architecture, but additionally includes a tokenization step that allows any kind of data to be learnt.

Variants

DSS has had a progression in versions and capability. Here you can explore their branches. DSS is an architecture for building content generation ML models using caches.

DSSv1

Character-by-character text generation using caches. 4 char keys, 1 char values. JSON-encoded.

DSSv2

Word-by-word text generation using caches. 1 word keys, 1 word values. JSON-encoded.

DSSv3

Word-by word text generation using caches. Variable word keys, 1 word values. JSON-encoded with brotli compression.

DSSv3g

The most capable DSS variant to date, capable of text generation, music generation, code completion and (limited) chatting.

General (integer tokens) version of DSS content generation using caches. Variable token keys, 1 token values. Binary encoded with brotli compression.

DSSv3i

Currently highly experimental and only implemented in JavaScript for easy graphics debugging.

Image generation using the DSS caches architecture. Variable-k scalar keys, pixel scalar values. Binary encoded with brotli compression.

Klarmuz

DSSv3g powers Klarmuz, the open platform for AI content generation. This project is possible thanks to jDev Studios and their fantastic Klarmuz compute units. For examples on running DSSv3g models, please check out the Klarmuz Platform.