Beat the long tail: Distribution-Aware Speculative Decoding for RL Training

Open in new window