CodeAct

Note

LLM agents are typically prompted to produce actions by generating JSON or text in a pre-defined format, which is usually limited by constrained action space and restricted flexibility. This work proposes to use executable Python code to consolidate LLM agents’ actions into a unified action space (CodeAct).

What is CodeAct?

In Fig. 2, we first introduce a general multi-turn interaction framework for LLM agents’ real-world usage that considers three roles: agent, user, and environment. CodeAct employs Python code to consolidate all actions for agent-environment interaction. In CodeAct, each emitted action to the environment is a piece of Python code, and the agent will receive outputs of code execution (e.g., results, errors) as observation.

CodeAct Shows the Promise as a Strong Tool Use Framework

Setup. We re-purpose API-Bank and test LLMs’ API-calling performance. For each evaluation instance, we instruct LLM to generate one atomic tool call in the format of a Python function call, JSON object, or text expression in a pre-defined format.

Results. For most LLMs, CodeAct achieves comparable or better performance even in atomic actions (the simplistic tool use scenario).

CodeAct Gets More Done with Fewer Interactions

M3ToolEval. To the best of our knowledge, no existing tool-use benchmarks contain complex tasks requiring the composition of multiple tools while supporting evaluating different action formats. Hence, we curate a benchmark M3ToolEval to fill this gap.

Setup. We allow the model to generate fully functional Python code that enables control and data flow (e.g., ifstatement, for-loop). Within each turn, the model can either emit an action or propose an answer to be verified by an exact match with the ground-truth solution. The interaction will terminate when a maximum of 10 interaction turns are reached or a correct solution has been submitted.

Results. CodeAct generally has a higher task success rate. Moreover, using CodeAct requires a lower average number of turns.

CodeAct Benefits from Multi-turn Interactions and Existing Software Packages

Thanks to its extensive knowledge of Python learned during pre-training, the LLM agent can automatically import the correct Python libraries to solve tasks without requiring user-provided tools or demonstrations.