Yale-BIDS-Chen/LEME-Benchmarks

LEME Open-ended Benchmarks

Task-specific configurations for the five open-ended benchmarks described in the LEME study. Available data are kept separate by task.

Study task	Study examples	Repository status
Patient QA (AAO)	76	Not distributed due to licensing restrictions
Public Eye Health QA (NEI/NIH)	281	Included: public_eye_health_qa
Patient Consultation (iCliniq)	303	Included: patient_consultation
Treatment Planning (EyeRounds)	245	Not distributed due to licensing restrictions
Abstract-to-Title Generation	500	Included: title_prediction

The three included configurations contain 1,084 examples. This is not a release of all 1,405 study examples. AAO and EyeRounds source texts are not included in the current release. availability.json records the status of all five tasks.

Loading

from datasets import load_dataset
dataset = load_dataset(
    "Yale-BIDS-Chen/LEME-Benchmarks",
    "patient_consultation",  # or public_eye_health_qa / title_prediction
    split="test",
)

Each configuration has the fields id, input, reference, and source. The title_prediction configuration also includes pmid, the PubMed identifier of the source article. Inputs and reference responses are unchanged from the original LEME archive. All splits are named test.

For source attribution, applicable source terms, and original-file hashes, see THIRD_PARTY_NOTICES.md and provenance.json.

Contact

For general inquiries, please contact Hyunjae Kim at hyunjae.kim@yale.edu.

License

LEME code, models, data, and other related materials may be used for non-commercial research purposes. Any derivative works must be made publicly available through a public platform, such as GitHub or Hugging Face. For commercial-use inquiries, please contact us (agreements@yale.edu or qingyu.chen@yale.edu). Please see the LICENSE file for the full terms.

The local LICENSE is an unmodified copy of the official GitHub license. As stated in that license, it applies only to rights held by the licensor; third-party materials remain subject to their applicable terms. Source attribution and applicable third-party notices are preserved separately.

Citation

@article{kim2026leme,
  title={LEME: open large language models for ophthalmology with advanced reasoning and clinical validation},
  author={Kim, Hyunjae and Ai, Xuguang and Srinivasan, Sahana and Gilson, Aidan and Singer, Maxwell B and Pushpanathan, Krithi and Xie, Qianqian and Park, Jungwoo and Applebaum, Serina and Yang, Gabriel Dawei and others},
  journal={npj Digital Medicine},
  year={2026},
  publisher={Nature Publishing Group UK, London}
}