Neura Market
Vicuna-13B logo

Vicuna-13B

Free

An open-source chatbot trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT. #opensource

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Company
LMSYS Org
Visit Website GitHub (40 stars)

About Vicuna-13B

Vicuna-13B is an open-source chatbot created by fine-tuning Meta's LLaMA model on approximately 70,000 user-shared conversations collected from ShareGPT. Developed by LMSYS Org, the model was trained on 8 A100 GPUs in one day at a cost of around $300. In preliminary evaluations using GPT-4 as a judge, Vicuna-13B achieved over 90% of the quality of ChatGPT and Google Bard, while outperforming other open-source models like Stanford Alpaca in most test cases. The code, weights, and an online demo are publicly available for non-commercial use, making it a cost-effective and accessible option for developers and researchers exploring conversational AI.

Key Features

Fine-tuned on 70,000 user-shared ChatGPT conversations from ShareGPT
Cost-effective training: ~$300 on 8 A100 GPUs in one day using PyTorch FSDP
Open-source code, weights, and online demo available for non-commercial use
Delivers detailed, well-structured responses comparable to ChatGPT
Scalable distributed serving system for handling multi-turn conversations
Outperforms LLaMA and Stanford Alpaca in GPT-4-assisted evaluations

Pros & Cons

Pros
  • Achieves ~90% of ChatGPT quality at a fraction of the cost
  • Fully open-source (code and weights) for non-commercial use
  • Low training cost of only $300
  • Provides detailed and well-structured answers
  • Capable of handling multi-turn conversations effectively
Cons
  • Evaluation is preliminary and non-rigorous (only 80 questions, GPT-4 as judge)
  • Limited to non-commercial use only
  • Based on LLaMA, which has its own usage restrictions
  • 13B parameter model may be resource-intensive for deployment
  • Not yet rigorously benchmarked against human judgment

Best For

Building conversational AI assistantsResearch in open-source chatbot development and evaluationPrototyping customer support bots or Q&A systemsStudying fine-tuning on user conversation data

FAQ

What is Vicuna-13B?
Vicuna-13B is an open-source chatbot created by fine-tuning Meta's LLaMA model on 70,000 user-shared conversations from ShareGPT.
How does Vicuna compare to ChatGPT?
In a preliminary evaluation using GPT-4 as a judge, Vicuna-13B achieved more than 90% of the quality of ChatGPT and Google Bard, and outperformed other open-source models like Stanford Alpaca.
Can I use Vicuna for commercial purposes?
No, the code and weights are released for non-commercial use only.
How much did it cost to train Vicuna?
Training Vicuna-13B cost approximately $300 using 8 A100 GPUs for one day.
Is Vicuna available to try online?
Yes, an online demo is publicly available to interact with the chatbot.