Overview

stanford-oval/Llama-2-7b-WikiChat is a 7 billion parameter LLaMA-2 model developed by Stanford OVAL, specifically fine-tuned to address the common issue of hallucination in large language model chatbots. The model's training incorporates WikiChat v1.0, a framework designed to ground chatbot responses on factual information extracted from Wikipedia.

Key Capabilities

• Hallucination Reduction: Significantly minimizes the generation of incorrect or fabricated information by grounding responses in verifiable sources.
• Wikipedia Grounding: Utilizes Wikipedia as a primary knowledge base to ensure factual accuracy in conversational outputs.
• LLaMA-2 Architecture: Built upon the robust LLaMA-2 (7B) foundation, providing strong language understanding and generation capabilities.

Use Cases

This model is particularly well-suited for applications where factual accuracy and the prevention of hallucination are critical. It can be used for:

• Factual Question Answering: Providing accurate answers to user queries by referencing Wikipedia.
• Information Retrieval Chatbots: Developing chatbots that can reliably summarize or extract information from a vast knowledge base.
• Educational Tools: Creating AI assistants that offer verifiable information for learning and research.

For more technical details, refer to the WikiChat GitHub repository and the associated research paper, "WikiChat: Stopping the Hallucination of Large Language Model Chatbots by Few-Shot Grounding on Wikipedia." An online demo is also available.