LongZero improves LLM long-context reasoning without external training data
Original title: LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data
Models write compact evidence, expand it into multi-document context, and train with GRPO on self-made tasks.
LongZero is a reinforcement-learning method that lets an LLM improve long-context reasoning without external corpora or a stronger teacher. The model first writes a question, answer, and compact supporting evidence, then expands that evidence into a long multi-document context. Those self-written multi-hop tasks are combined with self-written multi-round retrieval tasks and trained with GRPO. On two LLMs and five long-context benchmarks, the authors say LongZero beat RL that used data synthesized from external corpora and generalized to real tasks.
- Existing zero-data RL methods target short-context tasks.
- Training mixes multi-hop reasoning with multi-round retrieval tasks.
- Optimization uses GRPO.
- Evaluation covers two LLMs and five long-context reasoning benchmarks.
From the original article
1 min read at huggingface.co
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.
By Zhennan Wan, Jianfei Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.