How to Compare Open Large Language Models Using Chatbot Arena

If you are interested in the latest developments of open large language models (LLMs) for dialogue, you might want to check out Chatbot Arena, a platform that allows you to chat with different LLMs and vote for your favorite one. Chatbot Arena is hosted by LMSYS, a research group that develops Vicuna, an open-source chatbot that impresses GPT-4 with 90% ChatGPT quality.

LYMSYS - LLM Leaderboard


Chatbot Arena uses a simple and intuitive interface that lets you choose a model to chat with from a list of available options. You can also compare two models side by side or in a battle mode, where you chat with both models anonymously and then vote for the one that you think performed better. The votes are used to calculate the Elo ratings of the models, which are displayed on the leaderboard.


Vicuna vs FastChat


The leaderboard shows the ranking of the models based on their Elo ratings, as well as their descriptions and links to their websites or papers. You can see how different models compare to each other in terms of dialogue quality, personality, knowledge, and style. You can also see how Vicuna, the chat assistant fine-tuned from LLaMA on user-shared conversations by LMSYS, ranks among the top models.

Vicuna vs Dolly


Chatbot Arena is a great way to explore the capabilities and limitations of LLMs for dialogue, as well as to provide feedback and data for their improvement. You can also learn more about the research behind Chatbot Arena and Vicuna by reading their blog posts or joining their Discord community. If you are a researcher or a developer who wants to submit your own model to Chatbot Arena, you can contact LMSYS via email or Twitter.

Chatbot Arena is an open and fun platform that invites everyone to chat with open large language models and discover their potential and challenges. Whether you are a casual user or a serious researcher, you will find something interesting and useful in Chatbot Arena. So what are you waiting for? Go ahead and chat with some LLMs today here at - https://chat.lmsys.org/?leaderboard

No comments: