小模型也能高精度提取金融交易商户信息，省资源还快。
How Small Can You Go? LoRA Fine-Tuning 270M-8B Models for Merchant Information Extraction in Financial Transactions
- 用低秩适配（LoRA）微调270M~8B模型，平衡精度与效率。
- 0.8B的Qwen模型达94.75%准确率，4B版比8B版快3.8倍。
- 提示词设计影响大，选对模板能避免精度下降。
商户信息提取将嘈杂的金融交易描述转化为结构化字段，需在生产环境中实现高精度。我们部署的LoRA微调LLaMA-3.1-8B模型达到96.95% F1，但其内存与吞吐量限制了进一步应用。本文评估了23次微调实验及一个独立训练的生产基准，涵盖Gemma（270M–4B）、Qwen（0.8B–4B）、Aya（3.35B）和LLaMA-3.1-8B，覆盖不同LoRA秩、提示方式、训练模板与服务环境。其中，秩为8的LLaMA微调模型达96.75% F1，仅比秩32的生产参考低0.20点；而4B的Qwen使用JSON-Only提示达到96.60% F1与91.67%记录级精确匹配，逆吞吐时间估算仅为8B秩8模型的1/3.8。0.8B的Qwen达94.75% F1，Think与Nothink模板差异小于0.004 F1。在14个Databricks端点上，本地评估平均偏差-0.0081，唯独Aya家族出现2.7–5.1点下降。结果表明，紧凑型微调模型可保持近似精度，但模型选择需综合考虑提示策略、吞吐表现与服务栈行为。
原文摘要 · Abstract (English)
Merchant information extraction turns noisy financial transaction descriptors into structured fields at production scale. Our deployed LoRA-fine-tuned LLaMA~3.1-8B reaches 96.95\% F1, but its memory and throughput motivate smaller replacements. We evaluate 23 retained fine-tuning runs plus a separately trained production reference, spanning Gemma~3 (270M--4B), Qwen~3.5 (0.8B--4B), Aya~3.35B, and LLaMA~3.1-8B across LoRA ranks, prompts, training templates, and serving environments. A rank-8 LLaMA fine-tune reaches 96.75\% F1, only 0.20 points below the rank-32 production reference. Qwen~3.5~4B with JSON-Only prompting reaches 96.60\% F1 and strict record-level exact match of 91.67\%, with a $3.8\times$ lower inverse-throughput time estimate than the rank-8 8B model. Qwen~3.5~0.8B reaches 94.75\% F1, and Qwen Think and Nothink templates differ by less than 0.004 F1. Across 14 Databricks endpoints, mean F1 change from local evaluation is $-0.0081$; Aya is the only family with a 2.7--5.1 point decline. These results show that compact fine-tuned models can preserve most extraction accuracy, but model selection must account for prompt choice, throughput, and serving-stack behavior.
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。