-
I have a 4090. I thought I'd try the "reduce unet to float8" thing, but then python throws some kind of error like foreach_tensor_norm_cuda not implemented for float8 |
Beta Was this translation helpful? Give feedback.
Answered by
mx
May 5, 2024
Replies: 1 comment
-
Unfortunately, not really. Prodigy is just memory hungry. |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
mx
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Unfortunately, not really. Prodigy is just memory hungry.