Reinforcement Learning via Self-Distillation

Jonas Hübotter, Frederike Lübeck, Lejs Deen Behric, Anton Baumann, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, Andreas Krause

Keywords: einforcement learning, large language models, credit assignment, in-context learning, distillation, self-distillation, policy optimization, rich feedback

Abstract:

Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math.
Yet, current methods for reinforcement learning with verifiable rewards (RLVR) learn only from a scalar outcome reward per attempt, creating a severe credit-assignment bottleneck.
Many verifiable environments actually provide rich textual feedback, such as runtime errors or judge evaluations, that explain *why* an attempt failed.
We formalize this setting as reinforcement learning with rich feedback and introduce **Self-Distillation Policy Optimization** (SDPO), which converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model.
SDPO treats the current model conditioned on feedback as a self-teacher and distills its feedback-informed next-token predictions back into the policy.
In this way, SDPO leverages the model's ability to retrospectively identify its own mistakes in-context.
Across scientific reasoning, tool use, and competitive programming on LiveCodeBench v6, SDPO improves sample efficiency and final accuracy over strong RLVR baselines.
Notably, SDPO also outperforms baselines in standard RLVR environments that only return scalar feedback by using successful rollouts as implicit feedback for failed attempts.

Submission Number: 75