Please add a CPU and GPU multiprocessing feature to optimize the pre-processing (e.g. partitioning) using separate physical cores in parallel. It currently takes about 20-25 minutes to pre-process 700 textbook PDF pages in a single CPU core even though already following this for yolox. It would be excellent if you could implement multiprocessing with GPU. I can only create a workaround for the pdf file type because I do not know what method to split pages of other file types in chunks to many cores.