RL Post-training

conceptstub

RLHF, GRPO, and reward-model training loops.

Planned content

• Generation (inference) and training phases in one job
• Colocated vs disaggregated rollout engines
• Weight sync between engines