hybrid-reinforcement-learning

Combine sparse verifier rewards with dense reward model scores using stratified normalization to overcome limitations of either approach alone.

Independent GitHub snapshot · Automated classification · Original documentation governs