Support by Parquet, CSV, Jsonl, XLS
Compress text prompts efficiently
Compress prompts to speed up language model inference