StateFlow: Enhancing LLM Task-Solving through State-Driven Workflows

使用大型语言模型（LLMs）应对复杂任务的趋势是值得注意的，例如需要一系列动作和与工具和环境的动态交互的任务。本文提出了一种新型的基于LLM的任务解决范式——StateFlow，将由LLMs支持的复杂任务解决过程概念化为状态机。通过正确构建状态和定义状态转换，StateFlow实现了任务解决进程的基础，确保在整个任务解决过程中清晰跟踪和管理LLMs的响应。在每个状态中，StateFlow允许执行一系列动作，不仅涉及由特定提示指导的LLM响应的生成，还包括根据需要利用外部工具。状态转换由LLM做出的特定规则或决策控制，允许通过任务的预定义StateFlow模型进行动态和自适应的进展。对InterCode SQL和Bash基准测试的评估表明，StateFlow显著提高了LLMs的效率。

StateFlow: A State Machine-Based Paradigm for Large Language Model-Based Task Solving

StateFlow is a novel LLM-based task-solving paradigm that conceptualizes complex task-solving processes backed by LLMs as state machines, allowing for a dynamic and adaptive progression through the task's pre-defined StateFlow model.

StateFlow enhances LLMs' efficiency by grounding the progress of task-solving and ensuring clear tracking and management of LLMs' responses throughout the task-solving process. StateFlow allows execution of a series of actions, involving not only the generation of LLM's responses guided by a specific prompt, but also the utilization of external tools as needed. Evaluations on the InterCode SQL and Bash benchmarks show that StateFlow significantly enhances LLMs' efficiency.

Related work includes research on large language models, task-solving paradigms, and state machines. Some relevant papers include 'GPT-3: Language Models are Few-Shot Learners' by Brown et al., 'Task-Oriented Dialog Systems: A Survey' by Zhang et al., and 'A Survey of State Machine Replication Techniques' by Almeida et al.