Scientists from Berkeley, Stanford, and San Diego have introduced a new model, Vicuna, with 13 billion parameters (for comparison, GPT-3 has 175), which already surpasses its predecessor Llama and approaches the level of GPT-4, showing results at about ~90% of it! The most amazing part: you will be able to run this model on a regular desktop!
Training Vicuna costs only $300-500 on AWS, while training similar models from OpenAI costs around $10 billion.
Web: https://vicuna.lmsys.org/
Github: https://github.com/lm-sys/FastChat
Demo: https://chat.lmsys.org/
On the downside, commercial use is prohibited by the licensing agreement, but the first steps in OpenSource have been made.
