float16 when
#4
by Pinguin - opened
for colab users *sob
Changing the model accuracy is not required here.
What are you doing with your colab that is not supported?
This is for normal NVIDIA, colab is just T4, right? How is this a problem for you?
Original source is hunyuanvideo-community/HunyuanVideo , you can probably find what you need there.
We run this as a bfloat16.