Leo-Dai/dapo-math-17k_dedup
Viewer โข Updated โข 17.9k โข 16
Mid-Training Checkpoint of Parallel-R1: Towards Parallel Thinking via Reinforcement Learning
Stage: After 200 RL steps via alternating rewards โ showing the adaptive parallel reasoning ability and serve as structure exploration stage.
This checkpoint aims to help you reproduce experimental results in Section 4.5: Extra Bonus: Parallel Thinking as a Mid-Training Exploration Strategy for RL Training.