zhuifeng88
发表于 2023-4-20 12:31
fkpwolf 发表于 2023-4-20 12:17
https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/installation.html#prerequisites
...
这个只是库, 只有在hopper上才会用fp8专用单元加速, 在其他gpu上是软件优化
"Transformer Engine (TE) is a library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit floating point (FP8) precision on Hopper GPUs"
gtv
发表于 2023-4-20 15:01
代码库方便share吗?我这边有硬件,也想测测看
我輩樹である
发表于 2023-4-20 15:39
今天更新了cuda 12.1 release 1,根据nv开发者的信息transformer engine应该可以正常运作了。
https://github.com/NVIDIA/TransformerEngine/issues/15#issuecomment-1515703357