Post

The best embedding score available at the end of 2023 needed a 304M-parameter model. Today the same score comes from an 11M one — 28x fewer active parameters.

2 reposts15 likes