Writing

Longer pieces. Mostly post-training, evaluation that actually runs the output, and getting small models to punch above their parameter count. RSS.

Nothing here yet.