Self-Challenging Language Model Agents
This addresses the problem of reducing human annotation effort for training intelligent agents, though it is incremental as it builds on existing tool-use benchmarks.
The paper tackles the challenge of training language model agents for tool use by introducing a Self-Challenging framework where the agent generates its own high-quality tasks, called Code-as-Task, and trains on them with reinforcement learning. It achieves over a two-fold improvement in performance on benchmarks like M3ToolEval and TauBench using only self-generated data.
Large language models are quickly becoming the foundation for intelligent agents that are capable of using tools. However, training such agents is challenging because it requires human creation and annotation of a diverse set of tasks, tools, and evaluation criteria. In this paper, we propose the Self-Challenging framework for training an agent on high-quality tasks that are generated by itself. The agent first plays the role of challenger and generates a task after interacting with the given tools. The tasks take the form of a novel general class of problems termed Code-as-Task, which are defined by an instruction, a verification function and solution and failure cases which serve as tests, allowing to filter only for high-quality tasks. The agent then takes an executor role and trains on those tasks with reinforcement learning using the evaluation feedback as a reward. Evaluation on two existing multi-turn tool-use agent benchmarks, M3ToolEval and TauBench, shows the Self-Challenging framework achieves over a two-fold improvement in Llama-3.1-8B-Instruct, despite using only self-generated training data.