A public, reproducible retrieval benchmark for Indian automotive parts search — Hindi / Hinglish / English / misspellings / brand-as-generic / symptom / part-number query types. Built to let anyone independently verify claims about domain-tuned vs commercial embedding models on this task.

Companion to the model ManmohanBuildsProducts/auto-parts-search-v3 and the full eval writeup in docs/EVAL_REPORT.md.

	What's inside

Split File… See the full description on the dataset page: https://huggingface.co/datasets/ManmohanBuildsProducts/auto-parts-search-benchmark.