HN
Salomi, a research repo on extreme low-bit transformer quantization
SALOMI is a research repository focused on extreme low-bit transformer quantization and inference , especially the question of whether binary or near-binary weight representations can approach or exceed ternary baselines under realistic evaluation.
This repository is best treated as a research workspace rather than a one-command product package.
This is a research repository , not a polished production package.
The most important repo-level conclusion is:
Some materials under onebit/research/paper/ preserve earlier, more optimistic draft claims . For the most defensible current interpretation of the repository, prefer:
over historical paper-draft numbers when they conflict.