Magpie
by Magpie-Align · open-source · Last verified 2026-03-17
Magpie is a self-synthesis alignment dataset that harvests instruction-response pairs directly from aligned LLMs such as Llama-3-Instruct by feeding a pre-query template and letting the model generate both the user prompt and the assistant response. It produces over 3 million high-quality instruction-following samples without relying on seed tasks or manual curation.
https://huggingface.co/datasets/Magpie-Align/Magpie-Pro-1M-v0.1 ↗B
B—Above Average
Adoption: B+Quality: AFreshness: A+Citations: BEngagement: F
Specifications
- License
- Apache 2.0
- Pricing
- open-source
- Capabilities
- instruction-tuning, alignment, self-play
- Integrations
- huggingface-datasets
- Use Cases
- sft-training, rlhf-data, chat-model-finetuning
- API Available
- Yes
- Tags
- synthetic, instruction-tuning, alignment, self-synthesis, chat
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
62.1Adoption
74
Quality
80
Freshness
90
Citations
66
Engagement
0
Put AI to work for your business
Deploy this dataset alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.