Fetching the paper…

Training Ultra Long Context Language Model with Fully Pipelined Distributed Transformer · Around