arXiv cs.CL· Ken Enda, Jien Saito, Masami Mukai, Feng Han, Emir Turkes, Tatsumi Kusakabe, Reona Umeki, Kazuhiro Sumiyoshi, Mitsuya Numata, Takumi Tanikawa, Akira Endoh, Kenji Hirata, Atsushi Yoshida, Kohsuke Kudo, Shinya Tanaka·· 12 小时前AI 评分50
北海道大学医院在院内封闭网络部署 LLENS 大模型智能体平台
The LLENS Platform for On-Premises Large Language Model Agents in a Japanese University Hospital: Development and Evaluation Study
AI 导读
研究团队在北海道大学医院封闭网络内部署了 LLENS 临床支持智能体平台，单机 8 张 NVIDIA H200 GPU 运行约 1 万亿参数的开放权重模型 Kimi K2.6，配套基于 Open WebUI 重建的 Web 应用。
正文 · 原文
Authors:Ken Enda, Jien Saito, Masami Mukai, Feng Han, Emir Turkes, Tatsumi Kusakabe, Reona Umeki, Kazuhiro Sumiyoshi, Mitsuya Numata, Takumi Tanikawa, Akira Endoh, Kenji Hirata, Atsushi Yoshida, Kohsuke Kudo, Shinya Tanaka
Abstract:Large language model agents that use tools could take on part of the electronic health record and desk work that occupies physicians' time, but in many hospitals the medical information systems run inside closed networks, so the model and the agent harness must also sit inside. We designed, implemented, and deployed LLENS, a clinical support AI agent platform that runs inside the closed network of Hokkaido University Hospital on one host with 8 NVIDIA H200 GPUs, the open-weight model Kimi K2.6 (~1 trillion parameters), and a web application rebuilt from Open WebUI. The harness has 10 components, including page image reading, full-text attachment input, a path-addressed document tree without a vector index, a network-isolated sandbox, artifact return, self-checking of generated output, subagents, and saved procedures. Kimi K2.6 was chosen from 3 open-weight candidates and scored 480/500 on the 2026 Japanese Medical Licensing Examination; it generated 119 tokens/s for one user and served 16 concurrent requests at 908 tokens/s in total. On 180 questions over hospital documents (3 seeds each), mean accuracy was 10.7% with no documents, 81.3% with fixed retrieval (top 10 chunks of one lexical search), and 97.2% with interactive access, in which the model reads with document tools and the sandbox (15.9 percentage points above fixed retrieval, 95% CI 10.9-21.3); the difference lay in filtering/counting questions (16.7% vs 87.8%). Five scenarios with synthetic data ran on combinations of the components: needlestick injury response, batch adverse event grading, scanned referral letter transcription, falls data aggregation and plotting, and investigation of 3 topics by subagents. The platform has run on the hospital network since September 14, 2026.
| Comments: | 49 pages, 3 figures, 6 tables, with supplementary material. Submitted to JMIR Medical Informatics |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.32151 [cs.CL] |
| (or arXiv:2610.32151v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32151 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ken Enda [view email]
[v1]
Fri, 23 Oct 2026 04:34:26 UTC (1,861 KB)
来源：arXiv cs.CL · arxiv.org