... block that allows for attention and feedforward layers to be computed in parallel, enabling speedups from TPU compiler optimizations.
確定！ 回上一頁
