[论文] DeepEdu-v1: Efficient and Scalable Agentic LLMs for Vietnamese Educati...
研究领域: ML 作者: Quang Nguyen, Hieu Nguyen, Hien Hoang, Toan Pham, Cong Tran, Nam Vu 发布时间: 2026-09-25 arXiv: 2609.31568
论文概要
研究领域: ML 作者: Quang Nguyen, Hieu Nguyen, Hien Hoang, Toan Pham, Cong Tran, Nam Vu 发布时间: 2026-09-25 arXiv: 2609.31568
中文摘要
AI 辅导有望显著改善越南等发展中地区学生的学习成果,但两条显而易见的道路都有缺陷。ChatGPT 等云端助手把学生敏感数据路由到境外服务器,违反越南第 53 号法令等数据主权法律;且它们在以西方为中心的语料上预训练,并不围绕国家教科书课程组织,对本土内容的了解缺乏系统性、频繁产生幻觉。自行托管开源模型虽能把数据留在本地,却撞上双重墙壁:训练后量化(AWQ、GPTQ)虽然控制了静态权重占用,但长教学语境下的动态 KV 缓存与 prefill 延迟仍导致消费级 GPU 上的显存溢出和响应缓慢,且模型在地区特定材料上继续幻觉。我们提出 DeepEdu-v1——一个面向越南教育的 AI 辅导系统,构建在 SCALE(自改进上下文感知学习引擎)框架之上,该框架有两项创新。第一,长上下文推理引擎把 token 选择从每子块摊销到每聚簇粒度;在长上下文检索中,其检索调用次数比最先进的 selective-attention 基线少 7.7 倍,prefill 延迟(TTFT)降低约 35%,同时任务准确率持平或提升。第二,自改进智能体层不断从过往交互中整理经验证的操作手册,而不是微调——该设计旨在随着可信本地知识的积累,逐步降低对主导语言先验的依赖。在其部署配置下,DeepEdu 相比标准 vLLM 服务实现近 2 倍 TTFT 加速,并将复杂任务上的智能体准确率从 70.0% 提升至 79.5%,其中金融推理与交互式智能体基准的提升最为显著。
原文摘要
AI tutoring could markedly improve learning outcomes for students in developing regions such as Vietnam, yet the two obvious paths both fall short. Cloud assistants such as ChatGPT route sensitive student data to foreign servers---violating data-sovereignty laws such as Vietnam's Decree 53---and, pre-trained on Western-centric corpora, are not organized around the national textbook curriculum, so their knowledge of local content is unsystematic and frequently hallucinated. Self-hosting an open model keeps data on-premise but hits a two-fold wall: post-training quantization (AWQ, GPTQ) tames the static weight footprint, yet the dynamic KV cache and prefill latency of long tutoring contexts still cause out-of-memory failures and slow responses on consumer GPUs, while the model keeps hallucin...
*自动采集于 2026-09-29*
#论文 #arXiv #ML #小凯