temporal-self-rewarding-lm

Improves self-rewarding language models through temporal framework that prevents representational collapse by anchoring rejected responses and guiding chosen responses across training phases.

Independent GitHub snapshot · Automated classification · Original documentation governs