📄 中文摘要
AISA-AR-FunctionCall是一个面向生产的阿拉伯语函数调用框架，基于270M参数的FunctionGemma骨干网络构建。该框架通过系统的数据集审计、模式修复、工具感知提示重构和全参数监督微调进行训练。实验结果表明，在保留的测试集上，微调将解析失败率从87%降低到1%以下，函数名称准确率提高了八倍以上，并显著增强了跨方言和领域的参数对齐。错误分析显示，模型在处理阿拉伯语时实现了显著的结构稳定性，提升了其在多种应用场景中的可靠性。
从语言到行动：阿拉伯语中的可靠结构化工具调用通过数据中心微调
📄 English Summary
From Language to Action in Arabic: Reliable Structured Tool Calling via Data-Centric Fine-Tuning
AISA-AR-FunctionCall is a production-oriented Arabic function-calling framework built on a 270M-parameter FunctionGemma backbone. It is trained through systematic dataset auditing, schema repair, tool-aware prompt restructuring, and full-parameter supervised fine-tuning. Experimental results demonstrate that fine-tuning reduces parse failures from 87% to below 1% on a held-out test set, improves function name accuracy by over eightfold, and significantly enhances argument alignment across dialects and domains. Error analysis reveals a notable improvement in structural stability when handling Arabic, enhancing reliability across various application scenarios.