发表机构
Ertas AI（Ertas AI）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对QMSum无评分器问题，提出检索跨度训练方法，使406M模型在测试集上以36.33 ROUGE-1超越1.2B系统，且参数和内存更少，验证了该方法的有效性。
AI 中文摘要
QMSum未提供评分器，使得查询聚焦的会议摘要结果难以比较。我们在同一实现下对15个系统进行重新评分或生成。通过一个通用的推理端口，一个已发布的406M参数的Fusion-in-Decoder专家模型，在从截断的长输入切换到2000词的检索跨度时，ROUGE-1下降了6.30。在该跨度机制上对其进行微调可恢复这一损失。在测试集上，该模型得分为36.33 ROUGE-1，而我们的1.2B参数系统得分为35.41；会议聚类的95%置信区间为[-0.27, +2.22]，因此QMSum在统计上无法区分它们。较小的系统使用的总参数约为三分之一，峰值推理内存不到一半。在固定的1.2B基座内，跨度机制微调增加了5.29 [+4.02, +6.56]，而将前4500个转录词替换为2000个检索词在测试集上增加了1.55，在验证集上增加了0.29。另外，在一个简洁提示和参考重叠评分器下，一个已发布的406M专家模型在ROUGE-1上至少超过五个专有托管模型6.2，但输出长度以及缺乏人工或事实性评估限制了这一排序。结论仅限于QMSum和自动指标。
英文摘要
QMSum provides no scorer, making query-focused meeting summarization results difficult to compare. We rescore or generate 15 systems under one implementation. Through a common inference port, a released 406M Fusion-in-Decoder specialist loses 6.30 ROUGE-1 when moved from capped long input to 2,000-word retrieved spans. Fine-tuning it on this span regime recovers the loss. On test it scores 36.33 ROUGE-1 versus 35.41 for our 1.2B system; the meeting-cluster 95% interval for the difference is [-0.27, +2.22], so QMSum does not statistically separate them. The smaller system uses about one-third as many total parameters and less than half the peak inference memory. Within the fixed 1.2B base, span-regime fine-tuning adds 5.29 [+4.02, +6.56], while replacing the first 4,500 transcript words with 2,000 retrieved words adds 1.55 on test and 0.29 on validation. Separately, under one concise prompt and reference-overlap scorer, a released 406M specialist exceeds five proprietary hosted models by at least 6.2 ROUGE-1, but output length and absent human or factuality evaluation limit this ordering. Conclusions are limited to QMSum and automatic metrics.
Comments24 pages, 4 figures
