Medical QA Combined 10.5K: Instruction-Tuning Data for Clinical Language Models

Medical QA Combined (10.5k) is a dataset of 10,255 instruction-formatted medical question-answer pairs, created by nicholas-ugbala-hf. It combines 8,000 conversational patient-doctor exchanges from ChatDoctor HealthCareMagic with 4,000 encyclopedic clinical references from MedAlpaca WikiDoc. The dataset was last updated on June 12, 2026.

Use Cases

• Instruction fine-tuning of language models for medical dialogue based on conversational patient-doctor QA.
• Training clinical reference assistants based on encyclopedic clinical knowledge from WikiDoc.
• Evaluating model performance on medical QA tasks using the provided train/eval split of 9,229 and 1,026 rows.
• Cleaning and preprocessing text data by studying the applied regex methods for removing platform filler.

Strengths

• Contains 10,255 cleaned and filtered instruction examples.
• Combines two distinct medical QA sources: 8,000 conversational and 4,000 encyclopedic entries.
• Includes a predefined split of 9,229 training and 1,026 evaluation rows.
• Data is filtered to a maximum length of 512 tokens.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific cleaning process for removing platform filler is only partially described.
• Data may reflect source bias inherent to the original ChatDoctor HealthCareMagic and MedAlpaca WikiDoc collections.

Provenance

Source

nicholas-ugbala-hf

Collection Method

Cleaned combination and sampling from ChatDoctor HealthCareMagic and MedAlpaca WikiDoc.

Freshness

Last updated 2026-06-12 16:45:55; freshness should be verified.