L2MAC

L2MAC is a multi-agent generation framework that, a single input prompt can generate an extensive unbounded output, such as an entire codebase or an entire book. L2MAC can create near unbounded outputs that align exactly with the user input prompt over very long generation tasks It achieves strong empirical performance of state-of-the-art generation for large codebase tasks and is in the top 3 for the HumanEval coding global benchmark. As L2MAC can detect invalid code and failing unit tests when