StableVicuna

Overview

StableVicuna is the first large-scale open-source chatbot trained with reinforcement learning from human feedback (RLHF), launched by Stability AI, the company behind Stable Diffusion.

StableVicuna is a further instruction-tuned and RLHF-trained version of Vicuna v0.13b, which is an instruction-tuned LLaMA 13B model.

Core capabilities

Open-source accessibility

StableVicuna is released as an open-source chatbot, allowing researchers and developers to inspect, modify, and deploy it.

RLHF training

Trained with reinforcement learning from human feedback to align responses with human preferences.

Instruction-tuned

Further fine-tuned on top of Vicuna v0.13b to better follow user instructions.

Based on LLaMA 13B

Built on a 13-billion-parameter LLaMA model, providing strong language understanding and generation capabilities.

Online demo

Accessible through a hosted web interface at chat.lmsys.org for direct interaction.

Model selection

Users can choose StableVicuna from a dropdown menu in the demo interface.

Text interaction

Supports entering questions or instructions and receiving generated text responses.

Use cases

General-purpose conversational AI assistance

Answering knowledge-based questions

Following user instructions for task-oriented dialogue

Exploring open-source chatbot behavior for research

Prototyping applications that require a locally deployable chat model

Educational demonstrations of RLHF in language models

Evaluating model alignment with human preferences