Table 2 Statistics of our instruction-tuning datasets

From: Small language models learn enhanced reasoning skills from medical textbooks

Target application

Dataset

# Examples

Multiple-choice QA

MedQA-CoTa36

9308

 

MedBooks-18-CoTa

77,660

 

MedMCQA41

182,822

Free-form/Single-turn QA

LiveQA42

633

 

MedicationQA43

689

 

ChatDoctor-cleaneda38

111,902

Multi-turn QA

MedQA-dialoga36

4818

Clinical Note Generation

MTS-dialog60

1200

Miscellaneous

MedInstruct-52K44

52,002

  1. “# Examples” denotes the number of training examples for each dataset.
  2. aindicates that the dataset is newly constructed in this study. The total number of training examples is 441,034.
Back to article page