Mamabench: Normalized OBGYN and Pediatrics Question-Answering Benchmark

25,949 normalized rows form a benchmark for evaluating end-to-end medical question-answering systems. It was originally built to evaluate MAMAI, a RAG chatbot for nurses and midwives in Zanzibar. The dataset, authored by nmrenyi, includes multiple-choice and open-ended QA tracks.

Use Cases

• Benchmarking medical QA system performance based on normalized OBGYN and pediatrics questions.
• Evaluating retrieval-augmented generation (RAG) models based on the described reference architecture.
• Training or fine-tuning models for clinical advice chatbots based on the described use-case for nurses and midwives.

Strengths

• 25,949 normalized rows provide a substantial evaluation corpus.
• Includes multiple evaluation tracks: multiple-choice and open-ended QA with reference answers.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Likely curated from medical sources for benchmark construction.

Freshness

Last updated 2026-05-15 08:45:32; freshness should be verified.

Geography

Zanzibar (as the original use-case context)