RankVicuna: A Retrieval-Focused Chat Assistant

RankVicuna is a 7 billion parameter chat assistant developed by Castorini, built upon the Llama 2 architecture. It is fine-tuned from lmsys/vicuna-7b-v1.5 using supervised instruction fine-tuning on user-shared conversations collected from ShareGPT, with additional data augmentation. This specific model is provided in FP16 precision.

Key Capabilities

• Chat Assistant: Functions as a conversational AI, leveraging its training on diverse user interactions.
• Retrieval Research: Primarily intended for research applications combining large language models with information retrieval techniques.
• Llama 2 Base: Benefits from the foundational capabilities of the Llama 2 model family.

Good For

• Researchers: Ideal for those exploring the intersection of LLMs and retrieval, as detailed in its accompanying paper.
• Hobbyists: Suitable for individuals interested in natural language processing and information retrieval projects.
• Experimental Setups: Provides a fine-tuned model for evaluating and developing new approaches in these domains, with evaluations conducted on datasets like DL19/DL20.