metadata
license: apache-2.0
datasets:
- wikitext
The main stuff is stored in branches: BakaLLM is intented to be my pet project with series of gradual improvements. Basically one improvement per branch, from 001_pristine(simplest version) to ⑨⑨⑨_humanity_is_conquered.
HF repo contains code and weights. Github code only.
Code will not work AS IS, tinkering requirements. For one, I didn't place weight in weights dir(I wanted to zstd them first, but forgot) For two, bench (benchmark script to test against other models) is very opinionaned where models are stored.