EnterpriseSynth：从 OpenAPI 规范零执行生成 SFT 与评估数据
原标题：EnterpriseSynth: Zero-Execution SFT and Evaluation Data from OpenAPI Specs
AI 摘要
论文提出 EnterpriseSynth，一个零执行框架，可将 OpenAPI 规范转换为监督微调轨迹和配对评估实例，无需调用底层 API。该框架采用四阶段流水线（规范解析、意图生成、轨迹合成、静态验证），并在 17 个真实与合成 API 上评估。实验显示，对抗性精炼将验证器对植入模式违规的检测率从 57–80% 提升至 98%；在 EnterpriseSynth 数据上微调的小型开源模型在留出 API 上优于未调优模型和 Self-Instruct 基线。但独立 LLM 评估表明，端点级准确率（48.9%）会显著高估端到端正确性（21.3%），凸显了参数级评估的必要性。
正文节选
EnterpriseSynth: Zero-Execution SFT and Evaluation Data from OpenAPI Specs Abstract Tool-using large language model agents require substantial training and evaluation data, yet collecting execution traces from enterprise APIs is often unsafe, expensive, or infeasible because of access controls, compliance constraints, rate limits, and the risk of modifying production data. We present EnterpriseSynth, a zero-execution framework that converts OpenAPI specifications into supervised fine-tuning traj