> ML_DATASET // OPENASSISTANT-CONVERSATIONS_v1.0
OpenAssistant Conversations (OASST1)
LAION / Large-scale AI Open Network (Köpf et al. 2023) · Instruction Tuning & Conversational RLHF · 161,443 messages in 66,497 conversation trees across 35 languages (13,500+ volunteer contributors)
Instruction Tuning & Conversational RLHFApache-2.0161,443 messages in 66,497 conversation trees across 35 languages (13,500+ volunteer contributors)open
Dataset Profile & Characteristics
Label Type:Human ratings across quality, helpfulness, humor, toxicity, and role labels
Languages:en, es, de, fr, ru, zh, tr
License Tier:permissive-open-source
Modalities:text
Intended Use
- Instruction-tuning and reward model training for conversational open-weights assistants
Prohibited / Discouraged Use
- Medical diagnosis assistant training without clinical review
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
Crowdsourced volunteer contributions reviewed by community moderation for PII.
Known Bias:
Volunteer contributor demographic bias; diverse stylistic writing variations.
Known Benchmark Leakage:
Conversation trees strictly partitioned between train and validation sets.