Who am I?
A (brief) introduction to me - Romir! What I've done, am doing, and hope to do!
✦ INTERACTIVE · VISUAL CHAPTERS
EXPLORE THIS STORY ↗> cat /notes
FIELD
NOTES
Clear stories about curious ideas. Read them, play with the diagrams, and listen when a narrated version is available.
BLOG INDEX · 02 ENTRIES
LATEST
NOTE.
What DPO actually optimizes
Direct Preference Optimization turns a pair of preferred and rejected responses into a simple, useful training signal—without training a separate reward model.
✦ INTERACTIVE · VISUAL CHAPTERS
EXPLORE THIS STORY ↗