by LMSYS
Chatbot Arena Conversations is LMSYS's dataset of 33,000 cleaned conversations with pairwise human preference votes. It was collected on Chatbot Arena in 2023.
DatasetFree
Step 1: Sign in to Hugging Face and accept the dataset's conditions on its page.
Step 2: Install the datasets library and load the dataset from lmsys/chatbot_arena_conversations.
Step 3: Read the two model names, the conversations and the winner vote for each sample.
Step 4: Filter out unsafe conversations using the moderation and toxic tags before any training.
Step 5: Open the linked Colab notebook to see visualizations and Elo rating calculation.
User prompts are licensed CC-BY-4.0 and model outputs are CC-BY-NC-4.0.
It was collected from 13K unique IP addresses on Chatbot Arena between April and June 2023.
Yes. The authors kept unsafe conversations and tagged them. The card says it is not intended for training dialogue agents without appropriate filtering.
No. The terms say users agree not to attempt to determine the identity of individuals in the dataset.