Stable Vicuna
Stability AI Releases Vicuna, the First Open Source RLHF Chatbot
tag:AI Programming and DevelopmentAI training models StabilityAI training modelStableVicuna, the first large-scale open-source chatbot trained by Reinforcement Learning with Human Feedback (RHLF).StableVicuna is a further instruction-fine-tuned and RLHF-trained version of Vicuna v0 13b, which is an instruction-fine-tuned LLaMA 13b model.
Again, here are some benchmarks that show the overall performance of StableVicuna compared to other open source chatbots of similar size.
To realize the robust performance of StableVicuna, we utilize Vicuna as the base model and follow a typical three-stage RLHF pipeline as outlined by Steinnon et al. and Ouyang et al. The base Vicuna model was further trained by supervised fine-tuning (SFT) using three datasets:
data statistics
Data evaluation
This site AItools Artificial Intelligence Navigator website provides theStable VicunaAll from the network, does not guarantee the accuracy and completeness of external links, at the same time, for the pointing of this external link, not by the AItools Artificial Intelligence Navigation website actual control, at the time of inclusion in the July 16, 2024 am10:29, the content of the web page, all belong to the compliance and legal, the content of the later web pages, such as violations, you can directly contact the webmaster to delete. AItools Artificial Intelligence Navigator website does not assume any responsibility.