facebookincubator / AITemplate

AITemplate is a Python framework which renders neural network into high performance CUDA/HIP C++ code. Specialized for FP16 TensorCore (NVIDIA GPU) and MatrixCore (AMD GPU) inference.
Apache License 2.0
4.56k stars 370 forks source link

does aitemplate support fp32 stable diffusion? #125

Open xiayuqing0622 opened 1 year ago

xiayuqing0622 commented 1 year ago

does aitemplate support fp32 stable diffusion?

chenyang78 commented 1 year ago

Thanks for checking, @xiayuqing0622 !

No, we don't support fp32 stable diffusion in the current version. However, we just enabled fp32 support in the AIT core in our internal version. So, our next open source release will very likely contain fp32 support in general. We might be able to make our stable diffusion example support fp32 afterwards. Thanks.