TL;DR
This framework extends these paradigms by combining dynamic natural-language scaffolding with a non-monotonic decay schedule directly tied to stepwise RL sample rewards, and introduces an asymmetric Teacher-Student curriculum framework.
Extensive experiments show that Distilled RL substantially outperforms standard RL and OPD in terms of both pass@1 and pass@k, and can effectively transfer previously unavailable knowledge from a teacher model to a student model.
This work proposes a self-evolving curriculum that, each round, evaluates the current checkpoint, seeds QbQ from the problems it can mostly get right, and trains on the resulting variants; under an identical data budget, this breaks the ceiling and lifts pass@1 to 16.5\% with no sign of saturation after 20 rounds.
This thesis proposes a unified two-axis framework that organizes SFT and RL methods along a data axis (off-policy to on-policy) and a loss function axis (positive-only to positive-plus-negative to GRPO) and enables controlled ablations of individual components.
This work proposes $O^2-CritiCuRL, a novel curriculum reinforcement learning framework that introduces critical-step awareness through an iterative offline-online paradigm, and employs a progressive step-level reinforcement learning strategy, where truncated chains guide the model to infer missing steps and refine its...
LLM-enhanced automatic curriculum learning (LEACL), a framework that integrates LLMs and ACL to address limitations of ACL, is proposed, which achieves better asymptotic performance in terms of the success rates compared to human-designed dense rewards.
A systematic literature review on how RL are adapted and scaled as a fundamental post-training tools and how innovations in the RL pipeline enhance the domain-specific LLMs is conducted.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
