F1 training. #262
maybleMyers
started this conversation in
General
Replies: 2 comments 1 reply
-
|
Framepack? Have you tried testing the trained lora with fpack_generate_video specifically? The reason I ask is: |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
I only use musubi for inference. It looks like my LR was too low. I tried 4e-5 and it helped, it is working now. |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Has anyone had success F1 training? My loss looks good (0.046) but it is not learning any motion from the videos I am training on. I am trying 5e-5 LR w/ 128dim 64 alpha.
Beta Was this translation helpful? Give feedback.
All reactions