Compressing txtai embeddings
#5
by leok7v - opened
Compressing txtai embeddings index for the English edition of Wikipedia
Just FYI:
https://github.com/leok7v/wiki-slugs
nothing special but may help agents on a very small systems.
Very interesting, thank you for sharing!
I wonder if we can do similar thing to the sanitized (utf-8 text no markup) text of Wikipedia articles themselves)… I do not have compute resources for that but if you can do same as slugs for the e.g. executive summaries of the articles we can try hyper plane 1 bit compression on them too. May make reasoning on RAG for super small models super interesting…
Interesting. I'll think about this for the next update.