Crafting code-rich training data boosts AI models' long-context reasoning

Original: OctoLong: Mid-Training On Cross-Repository Code Contexts Enhances Long-Context Modeling

Source: arxiv.org ↗