Zhennan Wan, Jianfei Chen · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.30132 · cited by 0 works (OpenAlex) · open-access PDF
Усиление обучения в контексте (RL) является эффективным способом укрепления долгосрочного обоснования больших языковых моделей (LLM), но оно нуждается в данных обучения, которые являются длинными, разнообразными, сложными и проверяемыми. Человеческое обозначение таких данных дорогое, естественно возникающие данные слишком скудны, чтобы поддерживать RL, а существующие методы синтеза зависят от внешнего корпуса или более сильных моделей учителя, которые становятся труднее получить по мере того, как обученные модели становятся сильнее. В последнее время RL из нулевых данных, в которых модель учится от задач, которые она предлагает сама, достигла отличной производительности, но существующие методы направлены на краткосрочные задачи. Поэтому мы предлагаем $text\bfLongZero$, метод,
Machine translation (Meta M2M100); the English original follows.
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)
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.30132, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange