Wu Xing

PolicyLong advances LLM context extension with on-policy data evolution

Researchers have introduced PolicyLong, a novel method for extending the context windows of large language models by dynamically constructing training data. Unlike previous offline methods that use a fixed model to gene…