Nectar
by UC Berkeley · free · Last verified 2026-03-17
Nectar is a large-scale, high-quality preference dataset from Berkeley AI Research (BAIR). It contains 183,000 prompts, each with seven ranked responses from diverse models like GPT-4, ChatGPT, and open-source LLMs. It is designed for training robust reward models for RLHF and DPO.
https://huggingface.co/datasets/berkeley-nest/Nectar ↗B
B—Above Average
Adoption: BQuality: AFreshness: B+Citations: B+Engagement: F
Specifications
- License
- Apache-2.0
- Pricing
- free
- Capabilities
- reward-model-training, reinforcement-learning-from-human-feedback-rlhf, direct-preference-optimization-dpo, llm-alignment-and-safety, comparative-llm-evaluation, preference-learning-algorithms, instruction-following-research, ai-feedback-aif-pipelines
- Integrations
- Use Cases
- [object Object], [object Object], [object Object], [object Object], [object Object]
- API Available
- No
- Tags
- rlhf, preference-data, reward-model, llm-alignment, dpo, berkeley-bair, ai-feedback, ranked-responses, llm-evaluation, instruction-tuning
- Added
- 2026-03-17
- Completeness
- 0.85%
Index Score
61.6Adoption
66
Quality
86
Freshness
73
Citations
72
Engagement
0