renew-dev/nemotron-ab-persona-db-bge-m3

Nemotron-AB人物角色向量数据库（BAAI/bge-m3，19~59岁）是一个预构建的Chroma向量数据库，包含669,558个19至59岁的韩语人物角色（基于NVIDIA Nemotron-Personas-Korea数据集），使用BAAI/bge-m3模型进行嵌入（1024维，FP16精度，最大序列长度512）。该数据库以Chroma PersistentClient磁盘格式提供，可直接用于查询，支持韩语A/B人物角色模拟，例如在营销、UI文案、通知等场景中，通过LLM进行合成人物角色暴露评估。数据集包括集合marketing_personas，元数据字段如年龄、性别、职业、省份、地区、婚姻状况、教育水平、家庭类型、住房类型、兵役状态等，文档文本由多个属性连接而成。嵌入规格确保语义保留，使用时需使用相同模型和正则化。数据集总大小约11.2 GB，许可证为CC-BY-4.0，需注明原出处和修改事实。Nemotron-AB Persona VectorDB (BAAI/bge-m3, 19~59세) is a pre-built Chroma vector database of 669,558 Korean personas (from NVIDIA Nemotron-Personas-Korea dataset, ages 19–59) embedded with BAAI/bge-m3 (1024-d, fp16, max_seq_length=512). It is provided in Chroma PersistentClient disk format and ready-to-query for Korean A/B persona simulation, such as evaluating marketing, UI copy, or notifications by exposing synthetic personas to LLMs. The dataset includes the collection marketing_personas with metadata fields like age, sex, occupation, province, district, marital status, education level, family type, housing type, military status, and document text concatenated from multiple attributes. Embedding specifications ensure semantic preservation, requiring the same model and normalization for queries. The total size is approximately 11.2 GB, licensed under CC-BY-4.0, with attribution required to the original source and modifications. The Korean description from the README is: NVIDIA Nemotron-Personas-Korea 데이터셋에서 19~59세 한국인 페르소나 669,558건을 BAAI/bge-m3 로 임베딩해 Chroma PersistentClient 디스크 포맷으로 묶어 둔 사전 빌드 벡터 데이터베이스입니다. 마케팅·UI 카피·공지 등 한국어 단문/이미지 A/B 변형을 합성 페르소나에 노출시켜 LLM 으로 시뮬레이션 평가하는 nemotron-ab 프로젝트에서 즉시 검색에 사용할 수 있도록 만들어졌습니다.