fhai50032/latentsig-med-triage-router
收藏资源简介:
LatentSig医疗分诊路由器数据集是一个包含2000个经过验证的医疗分诊工具调用样本的数据集,其中1000个为英语样本,1000个为Hinglish(印地语和英语混合)样本,专门用于微调小型语言模型(SLMs)作为结构化医疗分诊路由器。该数据集旨在训练SLMs(1B-3B参数)成为可靠的临床医疗分诊结构化工具调用器。给定患者症状描述,模型必须:1. 从7个可用医疗工具中选择正确的工具;2. 输出带有正确参数的有效JSON工具调用;3. 分类紧急程度(急诊/紧急/半紧急/常规)。数据集设计原则是让模型学习跨工具泛化,而非记忆。每个样本都针对特定工具,确保覆盖整个医疗分诊领域。所有样本都经过三层验证流程(包括结构验证、语义检查和LLM法官评估)确保质量。数据集包含用户查询、响应、解析响应、工具调用、紧急类别、生成模型ID、语言、LLM法官ID、法官判决和哈希值等字段。
The LatentSig Medical Triage Router Dataset is a collection of 1,000 verified medical triage tool-call samples—500 English + 500 Hinglish—for fine-tuning Small Language Models (SLMs) as structured medical triage routers. This dataset trains SLMs (1B–3B parameters) to act as reliable structured tool-callers for clinical medical triage. Given a patient symptom description, the model must: 1. Select the correct tool from 7 available medical tools, 2. Output a valid JSON tool call with correct arguments, 3. Classify urgency (emergency / urgent / semi_urgent / routine). The key design principle is that the model learns to generalize across tools, not memorize. Each sample targets a specific tool, ensuring balanced coverage across the entire medical triage domain. All samples passed a 3-layer verification pipeline (including structural validation, semantic checks, and LLM judge evaluation) for quality assurance. The dataset includes fields such as user_query, response, parsed_response, tool_called, category, generation_model_id, language, llm_judge_id, judge_verdict, and hash.




