Prebuilt sorted indexes for the Kzr0xx/Icmr-and-hitek dataset (2.5B rows, 11 columns, ~104 GB raw parquet).
Building these indexes took ~17 hours of compute. This repo saves you that work: download + run = API live in ~1-2 hours (download speed dependent).
The indexes are stored as sorted parts (each < 50 GB, split at row-group boundaries, order preserved) because HuggingFace's classic HTTP… See the full description on the dataset page:
https://huggingface.co/datasets/Craige113/icrm-hitek-full-db-mixed.
