Newest papers

LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data

Zhennan Wan, Jianfei Chen · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30132 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Reforcement learning (RL) é uma maneira eficaz de fortalecer o raciocínio de longo contexto de grandes modelos de idiomas (LLMs), mas precisa de dados de treinamento que são longos, diversificados, desafiadores e verificáveis. A anotação humana de tais dados é cara, os dados que ocorrem naturalmente são escassos para sustentar o RL, e os métodos de síntese existentes dependem de corpos externos ou modelos de professores mais fortes, que se tornam mais difíceis de obter à medida que os modelos treinados se tornam mais fortes. Recentemente, o RL a partir de dados zero, em que um modelo aprende de tarefas que se propõe, alcançou um excelente desempenho, mas os métodos existentes visam tarefas de curto contexto. Portanto, propomos $textbfLongZ

Machine translation (Meta M2M100); the English original follows.

Abstract

Reinforcement learning (RL) is an effective way to strengthen the long-context reasoning of large language models (LLMs), but it needs training data that are long, diverse, challenging, and verifiable. Human annotation of such data is expensive, naturally occurring data are too scarce to sustain RL, and existing synthesis methods depend on external corpora or stronger teacher models, which become harder to obtain as the trained models get stronger. Recently, RL from zero data, in which a model learns from tasks it proposes itself, has achieved excellent performance, but existing methods target short-context tasks. Therefore, we propose $\textbf{LongZero}$, a method that lets an LLM improve its own long-context reasoning without any external data. Instead of writing questions and answers over existing documents, LongZero first has the model synthesize the question and answer with its supporting evidence in a compact structured form, and only then separately expands the evidence into a long multi-document context. LongZero combines these self-written multi-hop reasoning tasks with self-written multi-round retrieval tasks and trains the model with GRPO. We evaluate LongZero on two LLMs across five long-context reasoning benchmarks and compare it with RL on data synthesized from external corpora. Experimental results show that LongZero can generalize to real tasks, and outperforms both baselines, suggesting a viable and promising path to improving long-context reasoning of LLMs.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Artificial intelligence, Annotation, Path (computing), Reinforcement learning, Zero (linguistics)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange