We built it on Claude Sonnet 3.5 in early 2025. We upgraded to 3.7 without incident, and to 4.0 without incident. By the time ...
SDPG is the main contribution. It extends GRPO with an exact per-token forward KL between the actor (without privileged context) and itself conditioned on privileged context c: ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results