cs.LG提交于 2026-10-08已译

FinVector-Market-4B:面向结构化金融任务的 LoRA 适配受控研究

FinVector-Market-4B: A Controlled Study of LoRA Adaptation for Structured Financial Tasks

Alina Khaybullina

中文摘要

FinVector-Market-4B 在 22,000 条样本语料上,以 rank-16 LoRA 对 Qwen/Qwen3.5-4B 进行适配,用于结构化金融任务。同一份 600 条样本基准测试下,在隐式与显式 JSON Schema 契约条件下对基座模型与适配后模型进行评估。仅提供 Schema 即可将基座模型的 JSON 合法性从 0% 提升至 91.3%。在匹配显式提示下,冻结模型各项得分由 FinQA 答案精确匹配 14.7% 升至 40.0%,计算表达式正确性 48.0% 升至 82.7%,情景分支标签一致性 20.1% 升至 89.5%,隐含方向一致性 52.4% 升至 87.2%。事后策略打分审计显示,所报告的宏 F1 下降反映了标签集合的变化;在相同三类目标标签下,基座模型为 77.4%,适配模型为 83.1%。归档重叠与计算目标不一致削弱了基准泛化结论。结果表明,紧凑型金融领域适配在匹配提示下可在输出格式学习之外带来显著的任务特定收益,但收益受任务分布与提示契约约束。

关键要点

01问题:小模型仅做格式学习不够,需衡量 LoRA 适配在结构化金融任务上的真实增益
02方法:在 22,000 条样本上用 LoRA rank=16 适配 Qwen3.5-4B,对比隐式/显式 JSON Schema 提示
03结果:显式 Schema 提示下 JSON 合法性从 0% 升至 91.3%,FinQA 精确匹配 14.7%→40.0%,多类指标显著上升
04审计发现:宏 F1 下降源于标签集变化,同三类下基座 77.4% vs 适配 83.1%
05局限:归档重叠与计算目标不一致削弱基准泛化结论,收益受任务分布与提示契约约束

解读

尚无解读。

原始英文摘要

arXiv:2610.08882v1 Announce Type: new Abstract: FinVector-Market-4B adapts Qwen/Qwen3.5-4B with rank-16 LoRA on a 22,000-example corpus for structured financial tasks. We evaluate the base and adapted models on the same 600-example benchmark under implicit and explicit JSON-schema contracts. Supplying the schema alone raises base-model JSON validity from 0% to 91.3%. Under matched explicit prompting, the frozen scores improve from 14.7% to 40.0% for FinQA answer exact match, from 48.0% to 82.7% for calculator-expression correctness, from 20.1% to 89.5% for scenario branch-label agreement, and from 52.4% to 87.2% for implication-direction agreement. A post-hoc policy-scoring audit shows that the reported macro-F1 decline reflects a changing label set; using the same three target classes gives 77.4% for the base and 83.1% for the adapter. Filing overlap and calculator-target inconsistencies qualify the benchmark's generalization claims. The results show that compact financial domain adaptation can produce substantial task-specific gains beyond output-format learning under matched prompting, with gains bounded by the evaluated task distribution and prompt contract.