Rethinking Reinforcement Learning for Language Models: The SAO Approach
Single-rollout Asynchronous Optimization (SAO) offers a new path for more stable and effective reinforcement learning in large language models, challenging the traditional GRPO framework.
Read the full articleYou might also wanna read
SPIRAL: A Reinforcement Learning Framework for Multi-Primitive Language Model Reasoning
Language model reasoning can be substantially improved at test time via scaffolds that scale inference compute across different primitives -
arxiv.org·13d ago
ConSPO: A Contrastive Approach to Improving Reinforcement Learning with Verifiable Rewards for LLMs
Group Relative Policy Optimization (GRPO) is one of the most widely adopted RLVR algorithms for post-training large language models on reaso
arxiv.org·1mo ago
Understanding Reinforcement Learning for Model Training, and future directions with GRAPE
This paper provides a self-contained, from-scratch, exposition of key algorithms for instruction tuning of models: SFT, Rejection Sampling,
arxiv.org·9mo ago
Ouro: Looped Language Models That Build Reasoning into Pre-Training Through Latent Space Iteration
Modern LLMs are trained to "think" primarily via explicit text generation, such as chain-of-thought (CoT), which defers reasoning to post-tr
arxiv.org·6mo agoReinforcement Learning to Train Large Language Models to Explain Human Decisions
Article URL: Comments URL: Points: 6 # Comments: 0
Contextual Rollout Bandits: A Neural Scheduling Framework for Efficient Reinforcement Learning with Verifiable Rewards
Reinforcement Learning with Verifiable Rewards (RLVR) is an effective paradigm for improving the reasoning capabilities of large language mo
arxiv.org·1mo ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.