2,532 doctor-verified medical terms in 8 Indian languages, for testing embedding models.
Symptoms, diagnoses, procedures and medicines as people actually write them - misspelt, abbreviated, colloquial, in native or Latin script - each mapped by clinicians to a SNOMED CT concept. Queries, a 373,005-entry corpus and relevance labels.
Install
Download from Hugging Face
hf download ekacare/Eka-IndicMTEB --repo-type dataset Links
Topics
embeddingsretrievalbenchmarkmultilingualindicevaluation