Asclepius: Synthetic Clinical Notes and Instruction Dataset

157,000 synthetic discharge summaries generated from PMC-Patients case reports using GPT-3.5. The dataset is structured in a Clinical Note - Question - Answer format to build clinical LLMs. It was created by starmpcc and published on Hugging Face in June 2024.

Use Cases

• Training clinical question-answering models based on the Clinical Note - Question - Answer format
• Developing named entity recognition systems for medical text based on the supported tasks mentioned
• Generating synthetic training data for medical text summarization based on the discharge summaries
• Benchmarking instruction-following capabilities of LLMs in a clinical context based on the instruction-answer pairs

Strengths

• 157,000 synthetic discharge summaries provide a substantial corpus for training
• Dataset is structured specifically for clinical LLM development with note-question-answer format
• Based on PMC-Patients case reports, suggesting a foundation in real-world medical narratives

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown for the full dataset beyond the 157k discharge summaries
• Data is synthetic, generated by GPT-3.5, which may introduce artifacts

Provenance

Source

PMC-Patients case reports

Collection Method

Synthetic notes synthesized with GPT-3.5, followed by generation of instruction-answer pairs

Freshness

Last updated 2024-06-14 04:16:04