A balanced evaluation suite for LLMs and VLMs across 25 languages — all 22
scheduled Indian languages, plus Bhoti, Sinhala and English. Run it with
github.com/dhruvil237/indicbench.
uv run indicbench download --split main --out data/indicbench
uv run indicbench run --data data/indicbench
--backend vllm --base-url
http://localhost:8000/v1 --model
This is a personal, non-commercial research release by an individual, not an
organization's publication.… See the full description on the dataset page:
https://huggingface.co/datasets/dhruvil237/indicbench.
