Primary sourcearXiv cs.LG
View sourceEvent ·Multi-source
五款 7B-8B 开源模型土耳其语文档问答评测:本地 6GB 显卡跑出 49%-75% 准确率
一篇 arXiv 论文评测了五款 7B-8B 开源模型在土耳其语领域文档问答上的表现。主基准基于 109 页工业研发报告构建 100 道验证题,再用 112 页公共部门报告和独立 100 题集复现,全部在 6GB 显存的 NVIDIA RTX 3050 笔记本 GPU 上以 4-bit 量化本地运行。端到端准确率在 49% 到 75% 之间,评估协议通过证据标注把检索失败与模型推理失败分开。七种词法、稠密和混合检索配置经 McNemar 检验后,均未显著超过字符 TF-IDF 基线。
Current read
有人在 6GB 显存的笔记本上,用土耳其语 109 页报告实测了五款 7B-8B 开源模型,还顺手证明花哨检索方案打不过 TF-IDF,做 RAG 的可以看看。
8 sources41° AI heatLast updated 2026/9/23 12:33:57
Evidence
8 sourcesRelated source 1Thomas Wolf
View sourceRelated source 2NVIDIA AI
View sourceRelated source 3Rest of World
View sourceRelated source 4DAIR.AI
View sourceRelated source 5Clement Delangue
View sourceRelated source 6AIGCLINK
View sourceRelated source 7Teortaxes
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.