10 papers
Self-Improving Pretraining: using post-trained models to pretrain better models
Ellen Xiaoqing Tan, Jack Lanchantin, Shehzaad Dhuliawala +9
Large language models are classically trained in stages: pretraining on raw text followed by post-training for instruction following and reasoning. However, this separation creates…
Reasoning over mathematical objects: on-policy reward modeling and test time aggregation
Pranjal Aggarwal, Marjan Ghazvininejad, Seungone Kim +18
The ability to precisely derive mathematical objects is a core requirement for downstream STEM applications, including mathematics, physics, and chemistry, where reasoning must cul…
Hybrid Reinforcement: When Reward Is Sparse, It's Better to Be Dense
Leitian Tao, Ilia Kulikov, Swarnadeep Saha +5
Post-training for reasoning of large language models (LLMs) increasingly relies on verifiable rewards: deterministic checkers that provide 0-1 correctness signals. While reliable,…
J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning
Chenxi Whitehouse, Tianlu Wang, Ping Yu +4
The progress of AI is bottlenecked by the quality of evaluation, making powerful LLM-as-a-Judge models a core solution. The efficacy of these judges depends on their chain-of-thoug…
RESTRAIN: From Spurious Votes to Signals -- Self-Driven RL with Self-Penalization
Zhaoning Yu, Will Su, Leitian Tao +9
Reinforcement learning with human-annotated data has boosted chain-of-thought reasoning in large reasoning models, but these gains come at high costs in labeled data while falterin…
CoT-Self-Instruct: Building high-quality synthetic prompts for reasoning and non-reasoning tasks
Ping Yu, Jack Lanchantin, Tianlu Wang +6
We propose CoT-Self-Instruct, a synthetic data generation method that instructs LLMs to first reason and plan via Chain-of-Thought (CoT) based on given seed tasks, and then generat…