> ML_DATASET // LMSYS-CHAT-1M_v1.0
LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset
LMSYS Org / UC Berkeley (Zheng et al.) · Conversational AI & User Intent · 1 million real-world user conversations with 25 frontier LLMs collected from Chatbot Arena
Conversational AI & User IntentLMSYS Dataset License (Non-Commercial Research)1 million real-world user conversations with 25 frontier LLMs collected from Chatbot Arenaopen
Dataset Profile & Characteristics
Label Type:Multi-turn user prompts and model responses with safety tags and language metadata
Languages:en, es, zh, ru, de, fr, tr, ja
License Tier:non-commercial
Modalities:text
Intended Use
- Analyzing real-world user intent distributions, instruction tuning, and conversational clustering
Prohibited / Discouraged Use
- Re-identifying individual Chatbot Arena participants
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
Extensive regex and model-based filtering applied to scrub user PII and toxic submissions.
Known Bias:
Chatbot Arena user demographic bias; tech-savvy curiosity queries.
Known Benchmark Leakage:
Real user test queries may unintentionally probe live production endpoints.