Miriad 5.8M: Medical Instruction and Retrieval QA Pairs from Literature

Miriad 5.8M contains 5.8 million medical question-answer pairs distilled from peer-reviewed biomedical literature using Large Language Models. Released in June 2025 by the Miriad research team, the dataset provides structured data for medical instruction tuning and retrieval-augmented generation. It serves as a large-scale resource for training models on verified scientific knowledge rather than general web content.

Use Cases

• Instruction tuning of medical LLMs using the question-answer pairs
• Benchmarking Retrieval-Augmented Generation (RAG) systems on biomedical facts
• Hallucination detection by verifying model responses against literature-derived QA pairs

Strengths

• 5.8 million record count
• Sourced from peer-reviewed biomedical literature
• Structured specifically for instruction tuning and RAG tasks

Limitations

• LLM-distilled synthetic labels may contain distillation bias or hallucinations from the teacher model
• Lack of raw source text context for some pairs may limit full document retrieval testing

Provenance

Source

Peer-reviewed biomedical literature

Collection Method

LLM-based distillation

Freshness

Last updated June 2025; based on literature available up to the distillation date.