warpt-ctc

warp-ctc is an open-source, high-performance library that implements Connectionist Temporal Classification (CTC) loss calculation for sequence learning tasks such as speech recognition. Developed by Baidu Research, it provides parallelized CTC computation on both CPU and GPU (CUDA), significantly improving training speed and scalability for deep learning models. It is especially useful for research labs and developers building and training large-scale end-to-end sequence models where alignment-free learning is required.