CASE EVIDENCE / A RECORD
Team5_NLP_Upstage project team 公开的代码审查与测试案例，来源为 公开代码库，复核于 2026-06-27T08:06:43Z。
原始记录：Team5_NLP_Upstage built a RAG and prompt-routing QA pipeline with Solar Pro 2
Chinese Brief
Team5_NLP_Upstage project team 公开的代码审查与测试案例，来源为 公开代码库，复核于 2026-06-27T08:06:43Z。 Model Atlas 将它标记为 A 类证据，因为它同时具备具体使用者、具体任务、公开原始证据和可访问产物。 Model Atlas 不把 benchmark、教程、发布说明或集合页包装成真实案例。
任务
这是一个围绕代码审查和测试生成的真实任务，公开材料可以回溯到具体使用者和具体产物。 原始资料写作：The Team5_NLP_Upstage term project uses Solar Pro 2 with RAG, domain routing, prompt engineering, table parsing, and majority-vote ensembling to answer EWHA and MMLU-style questions from supplied datasets.
公开材料提供公开代码、README 或项目配置，可用于核验任务结果、项目形态和模型绑定关系。 原始资料写作：The repository reports 88% total performance on 5_final.csv, including 100% on EWHA questions and 76% on MMLU questions, and provides runnable code that writes answers to 5_final.csv.
Solar Pro 2 在该案例中承担代码审查和测试生成相关的生成、分析、编排或实现角色。 原始资料写作：Solar Pro 2 is instantiated through LangChain ChatUpstage for both EWHA voting and MMLU domain-specific answering, while Upstage embeddings retrieve context; the model generates the final answer choices after routing an…
当前判断基于公开材料；若产物下线、仓库变更或模型参与比例仅来自作者自述，需要在引用前重新复核。 原始资料写作：Academic project with evaluation-style output, not a commercial deployment; accepted because the repo contains a concrete runnable QA application with public code, task, output, and explicit solar-pro2 model usage rathe…