AgentPublic/fabrique-reference-2
AgentPublic/fabrique-reference-2 is a 13 billion parameter language model designed by AgentPublic, featuring a 4096-token context length. This model is specifically engineered for generating responses within the servicespublics+ ecosystem, offering distinct modes for experience-based content creation and expert document-driven information enrichment. It excels at integrating RAG capabilities to reduce hallucinations in contact data and links, making it suitable for structured content generation in public service applications.
Loading preview...
Model Overview
AgentPublic/fabrique-reference-2 is a 13 billion parameter language model developed by AgentPublic, tailored for generating structured responses within the servicespublics+ framework. It operates with a 4096-token context length and integrates Retrieval Augmented Generation (RAG) across its operational modes to enhance factual accuracy and relevance.
Key Capabilities
- Simple Mode: Functions as the original etalab model (fabrique-miaou), generating answers from existing experiences, answer elements, and links.
- Experience Mode: Utilizes RAG on servicespublics+ experiences, searching for similar institutional experiences and integrating them into the prompt. This mode aims to minimize hallucinations related to contact data and links by leveraging existing, relevant responses.
- Expert Mode: Employs RAG on expert documents from servicespublics.fr or other reference sources. It enriches pre-generated responses with advanced, standardized information and references, following a Wikipedia-like model with tags.
Use Cases
This model is particularly well-suited for applications requiring the generation of public service-oriented content. Its multi-mode design allows for a two-stage user journey:
- Initial Response Generation: Using Simple or Experience mode to craft the first response to an experience in the servicespublics+ style.
- Information Enrichment: Applying Expert mode to add referenced, advanced information to the initial response, ensuring accuracy and standardization.