用纠错修订法让小模型超越大模型的客服摘要效果
Error-Aware Knowledge Distillation via Targeted Revision for Customer-Service Summarization
- 分析大模型错误并用小编辑模型精准修正数据
- 小模型在修正数据上训练后性能超GPT-3.5
- 适合追求低成本高隐私的工业级应用
我们提出分析-修订-微调（ARF）流程，使小型开源语言模型在客服摘要任务中显著超越大型专有模型。该流程首先分析并分类教师模型（GPT-3.5）生成摘要中的常见错误，随后利用紧凑的编辑模型（Llama 3.1 70B）进行针对性修订，生成高质量、精细化的训练数据。将小型学生模型（如 Llama 3.1 8B、QWen3 4B）在此类数据上微调后，其摘要性能优于 GPT-3.5。ARF 流程在提升成本效益与数据隐私保护的同时，保持了竞争力强的准确性，展示了一种可泛化应用于多种下游任务的开源大模型增强框架。
原文摘要 · Abstract (English)
We introduce an Analyze-Revise-Finetune (ARF) pipeline that enables smaller open-source language models (LLMs) to surpass substantially larger proprietary models in customer service summarization tasks. The pipeline first analyzes and categorizes common errors in summaries produced by a teacher model (GPT-3.5), then performs a targeted revision using a compact editor model (Llama 3.1 70B) to generate high-quality, refined training data. Fine-tuning smaller student models (e.g., Llama 3.1 8B, QWen3 4B) on this refined data resulted in superior summarization performance compared to GPT-3.5. The ARF pipeline improves cost efficiency and data privacy while maintaining competitive accuracy, illustrating a generalizable framework for enhancing open-source LLMs across diverse downstream applications.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。