This fork make sentence splitting language agnostic by introducing wtpsplit and 500M (for onnx) runtime download size. googletrans and sentence-splitter is removed. This project has full language agnostic now since LaBSE is already language agnostic.
This fork is willing to support more embedding models, such as harrier-oss, which is current bitext SOTA on MMTEB. However LaBSE is still very high ranked on MMTEB though.