pipe.enable_attention_slicing()
#11
by
Teetle44
- opened
As mentioned on the model card page: "If you have low GPU RAM available, make sure to add a pipe.enable_attention_slicing() after sending it to cuda for less VRAM usage (to the cost of speed)"
Which file do I add the pipe.enable_attention_slicing() to?
I think this might help you
https://github.com/huggingface/diffusers/commit/d50e3217459558cc2979f38818f1835751d4fc97
pipe = pipe.to("cuda")
pipe.enable_attention_slicing()
Thanks!