Overview
glaiveai/Llama-3-8B-RAG-v1 is an 8 billion parameter model developed by Glaive, built upon the Llama-3 architecture. It has been specifically fine-tuned for Retrieval-Augmented Generation (RAG) tasks using the Glaive-RAG-v1 dataset. This model is designed to provide answers based on provided documents, with a strong emphasis on factual grounding and citation.
Key Capabilities
- Retrieval-Augmented Generation (RAG): Optimized to generate responses by leveraging information from external documents.
- Answer Modes: Supports two distinct answer modes:
- Grounded: Responds strictly with facts extracted from the provided documents.
- Mixed: Combines facts from documents with the model's internal knowledge.
- Citation Generation: Automatically generates citations, both at the beginning of the answer and inline within the text using
<co: doc_id></co> tags, linking information directly to its source document. - Structured Input: Designed to work with a specific input format, including document content, titles, and user queries, to ensure optimal performance.
Use Cases
This model is particularly well-suited for applications requiring accurate, verifiable, and source-attributable responses. Ideal scenarios include:
- Grounded Question Answering: Answering user queries based on a provided corpus of documents.
- Information Extraction: Extracting specific facts and details from texts with clear source attribution.
- Knowledge Base Interaction: Building conversational agents that can reference and cite information from a knowledge base.
Developers should adhere to the specified system prompt and document formatting to achieve the best results, as this aligns with the model's fine-tuning methodology.