HiMed

# HiMed HiMed is a Hindi medical dataset and benchmark suite covering both Western medicine and Indian systems of medicine. It consists of two parts: - **HiMed-Trad**: traditional Indian medicine - **HiMed-West**: Western medicine under Hindi prompts ## Repository Layout All released files are under `data/`: ```text data/ ├── HiMed-Trad_Bench.json ├── HiMed-Trad_Corpus.json ├── HiMed-West_Bench.json ├── HiMed-West_Corpus.json └── HiMed-West_Exam.json ``` > We define multiple Hugging Face dataset configurations (`configs`) because these JSON files do not share an identical schema, and the dataset viewer would otherwise try to merge them into a single table. ## Dataset Configurations Load each subset via `config_name`: - `himed_trad_corpus` - `himed_trad_bench` - `himed_west_corpus` - `himed_west_bench` - `himed_west_exam` ## Statistics The number of items in each JSON file matches the counts below: | Subset | Size | |---|---:| | HiMed-Trad Bench | 6,010 | | HiMed-West Bench | 1,784 | | HiMed-West Exam | 470 | | HiMed-Trad Corpus (full) | 286,657 | | HiMed-West Corpus (full) | 116,859 | ## Quickstart ```python from datasets import load_dataset bench = load_dataset("FreedomIntelligence/HiMed", "himed_trad_bench", split="test") print(len(bench), bench[0]) corpus = load_dataset("FreedomIntelligence/HiMed", "himed_trad_corpus", split="train") print(len(corpus), corpus[0]) ``` ## License Released under the **Apache License 2.0** (see `LICENSE`).