Medical Reasoning Corpus · miriad
miriad-5.8M
A dataset of 5.8 million biomedical question-answer pairs for training and evaluating healthcare language models.
What miriad-5.8M is
MIRIAD (Medical Instruction and RetrIeval Dataset) is a corpus of 5.8 million medical question-answer pairs designed to support healthcare language model development. Built by Qinyue Zheng, Salman Abdullah, Sam Rawal, Cyril Zakka, Sophie Ostmeier, Maximilian Purk, Eduardo Reis, Eric J. Topol, Jure Leskovec, and Michael Moor, it provides large-scale data for clinical reasoning and instruction-based learning.
The corpus addresses the demand for structured, high-volume biomedical dialogue grounded in scientific publications. It serves researchers developing and evaluating models for retrieval-augmented generation, information retrieval, hallucination detection, and medical instruction tuning.
How it was made
The dataset was created using biomedical literature from the Semantic Scholar Open Research Corpus (S2ORC). Articles from this corpus were supplied as inputs to OpenAI language models, which distilled the scientific texts into structured question-answer pairs focused on medical knowledge.
Key facts
- Total examples
- 5.8 million question-answer pairs
- Source corpus
- Semantic Scholar Open Research Corpus (S2ORC)
- Distillation model
- OpenAI language models
- Data format
- Parquet
- Modality
- Text
- Publication year
- 2025
What you can use it for
- Instruction tuning
- Fine-tuning healthcare-specific language models on clinical and biomedical question-answering tasks.
- Retrieval-augmented generation
- Supplying paired queries and passages to train and evaluate medical retrieval-augmented generation systems.
- Hallucination detection
- Evaluating whether clinical language models produce ungrounded or factually incorrect claims.
- Biomedical information retrieval
- Developing and testing document retrieval engines tailored to peer-reviewed biomedical literature.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.How the columns are distributed
Over 2,092,264 rows of default/train
answer
string_text- min
- 17
- median
- 474
- mean
- 488.97
- max
- 2,490
paper_id
string_text- min
- 3
- median
- 8
- mean
- 7.79
- max
- 9
paper_title
string_text- min
- 5
- median
- 97
- mean
- 100.78
- max
- 300
paper_url
string_text- min
- 44
- median
- 49
- mean
- 48.79
- max
- 50
passage_position
string_label- 032%
- 119%
- 215%
passage_text
string_text- min
- 103
- median
- 4,510
- mean
- 4558.46
- max
- 9,421
qa_id
string_text- min
- 9
- median
- 15
- mean
- 14.61
- max
- 17
question
string_text- min
- 19
- median
- 100
- mean
- 106.20
- max
- 496
specialty
string_label- Public Health & Epidemiology17%
- Basic Medical Sciences12%
- Pharmacy & Pharmacology8%
venue
string_text- min
- 1
- median
- 28
- mean
- 31.89
- max
- 300
0.43% empty
year
float- min
- 1,793
- median
- 2,014
- mean
- 2010.94
- max
- 2,021
0.01% empty
Strengths and limits
Good at
- Provides 5.8 million structured question-answer pairs derived from peer-reviewed scientific literature.
- Supports multiple distinct natural language tasks, including retrieval, reasoning, and instruction tuning.
Watch out for
- Not evaluated or approved by regulatory bodies, preventing its use as a medical device.
- Unsuitable for clinical decision-making or direct medical diagnosis concerning real individuals.
- Contains synthetic outputs distilled by language models, which may reflect generation errors.
Questions
Can I look inside miriad-5.8M without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is miriad-5.8M under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
Where does the underlying literature originate?
The text inputs are derived from biomedical publications indexed in the Semantic Scholar Open Research Corpus (S2ORC).
Can this dataset be used to inform real-world patient diagnoses?
No. The dataset is restricted to academic research and educational exploration, and must not be used for diagnosis or treatment decisions.
How were the question-answer pairs produced?
They were distilled from S2ORC scientific papers using OpenAI language models to generate structured questions and answers.
How to cite it
Qinyue Zheng, Salman Abdullah, Sam Rawal, Cyril Zakka, Sophie Ostmeier, Maximilian Purk, Eduardo Reis, Eric J. Topol, Jure Leskovec, Michael Moor. MIRIAD: Augmenting LLMs with millions of medical query-response pairs. arXiv:2506.06091, 2025.
Written from miriad-5.8M's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.