Vicuna 13B

Vicuna-13B is a finely-tuned open-source chatbot derived from the LLaMA model and developed with around 70,000 user-shared conversations from ShareGPT. Built on the robust Transformer architecture, it features a substantial 13-billion parameter scale. Early evaluations indicate it achieves over 90% of the effectiveness of models like OpenAI's ChatGPT and Google's Bard, surpassing other open-source models such as LLaMA and Stanford Alpaca in various scenarios. Training data includes user conversations initially captured in HTML and converted to markdown for quality filtering. Noteworthy advancements include memory optimizations allowing a context length of 2048 and enhanced multi-turn conversation handling, although it faces challenges in reasoning, mathematics, and factual consistency. It lacks complete optimization for safety and bias reduction. Available in several versions, including a 4-bit quantized edition for efficiency, Vicuna-13B is accessible for non-commercial application.

Vicuna 13B is an open-weight model in the Vicuna family. The structured metadata tracks a 2k-token context window and structured outputs. This page tracks provider routes through GCP Vertex AI and Replicate API, with the cheapest tracked route listed at $0.1 input and $0.5 output per 1M tokens. No headline benchmark score is tracked for Vicuna 13B yet.