PipelineRL: Faster On-policy Reinforcement Learning for Long Sequence Generation

Rafael Pardinas, Ehsan Kamalloo, Alexandre Piche, Dzmitry Bahdanau

janvier 2026