[QA]   Block Transformer: Global-to-Local Language Modeling for Fast Inference | Arxiv Papers | Podwise