Vicuna

Vicuna

Vicuna 13b, FastChat alpaca model, based on Meta's open-source LLaMA

Visit Site →

Vicuna Official Website 13b, FastChat Alpaca Model, Based on Meta's Open Source LLaMA Large Alpaca

What is Vicuna?

Open-sourced by MetaLLaMAStarting with the LLaMA series models (translated as "Large Alpaca"), researchers have gradually developed lightweight ChatGPT-like models based on LLaMA, such as Alpaca, Alpaca-Lora, and Luotuo (Camel), and opened them up. Recently, researchers have proposed a new model: Vicuna (Alpaca). This model is based on LLaMA with 13B parameters. Vicuna-13B achieves high-performance dialogue generation through fine-tuning Q LLaMA. Interestingly, during the model evaluation, the authors did not use a "standardized test" to measure model performance but had GPT-4 act as the examiner to see which answers GPT-4 prefers from Vicuna-13B or other baseline Q models.

Vicuna Official Website:https://github.com/lm-sys/FastChat
Demo Address:https://chat.Imsys.org/

The results show that compared to existing SOTA open-source models (LLaMA, Alpaca), GPT-4 prefers Vicuna in over 90% of the questions, and Vicuna achieves 92% of the total score of ChatGPT.Vicuna (Alpaca) is now open-source