ecnote-outputs / README.md
Sakhil's picture
Outputs as one archive (ecnote-outputs.tar.gz) instead of loose files; card updated
9ff756b verified
|
Raw History Blame Contribute Delete
3.21 kB
metadata
license: cc-by-4.0
language:
  - en
pretty_name: Clinical Note Generation - Run Outputs (PriMock57, ACI-Bench)
tags:
  - medical
  - clinical-notes
  - doctor-patient-conversations
  - llm-generated

Clinical Note Generation: Run Outputs

Machine-generated outputs of the ClinicalNoteGeneration project (team Empty-NLP): clinical notes written from doctor-patient conversations, their automatic evaluation records, run logs and model response caches. Code: github.com/tanejasakhil/ClinicalNoteGeneration.

Source data and licence

Everything here is generated from the conversations and reference notes of these public datasets:

Dataset Licence Reference
PriMock57 CC BY 4.0 Korfiatis et al., PriMock57: A Dataset of Primary Care Mock Consultations, ACL 2022
ACI-Bench CC BY 4.0 Yim et al., ACI-BENCH: a Novel Ambient Clinical Intelligence Dataset for Benchmarking Automatic Visit Note Generation, Scientific Data 2023
MTS-Dialog (small part, in mts/ and validation/) CC BY 4.0 Ben Abacha et al., An Empirical Study of Clinical Note Generation from Doctor-Patient Encounters, EACL 2023

As derived material, these outputs are released under the same licence, Creative Commons Attribution 4.0 International (CC BY 4.0). If you use them, please credit the datasets above as well as this project. The conversations are mock or role-played consultations (as described by the dataset authors); no real patient data is included.

Text produced by a model may also be subject to that model's own licence or terms of use.

Models

Role Models
Note generation Qwen3.8-27B (main runs), Qwen3.6-35B-A3B (earlier runs)
Other models in small comparison runs Qwen3.5-9B, HuatuoGPT-3
Judging Muse-Glimmer-30B (main judge); Gemma-4-31B and MedGemma-27B in small judge comparisons
Automatic text scores DeBERTa-v3-large (MNLI/FEVER/ANLI/LING/WANLI), RoBERTa-large, DeBERTa-xlarge-MNLI, BioClinical-ModernBERT, BLEURT-20

All models were self-hosted (GGUF builds served with llama.cpp for the LLMs); no hosted API was used.

Files

File Size SHA-256
ecnote-outputs.tar.gz 294 MB (1.6 GB unpacked, about 32,400 files) 30c9038771c24adfc66a474fd7ca6a4a6dfa5d17e0bbbd9da3a1e6ac5cfc5fd7

The archive holds one folder, outputs/: generated case sheets, judge records, text metrics and run logs per dataset (primock57/, aci/, mts/, mini/), the model response cache (llm_cache/, llm_log.jsonl), the NLI cache (nli_cache.sqlite) and validation outputs. Earlier commits of this repo hold the same outputs as loose files.

Use

Run in the root of the code repository, so the archive unpacks into its outputs/ folder:

hf download Sakhil/ecnote-outputs ecnote-outputs.tar.gz --repo-type dataset --local-dir .
sha256sum ecnote-outputs.tar.gz      # optional: compare with the table above
tar -xzf ecnote-outputs.tar.gz && rm ecnote-outputs.tar.gz

Notes are machine-generated and have not been reviewed by clinicians; they are for research only and must not be used for patient care.