LoRA tab query
#103
-
|
What does the LoRA base model do? |
Beta Was this translation helpful? Give feedback.
Answered by
O-J1
Oct 21, 2024
Replies: 1 comment
-
|
Please read the wiki. Leave values at defaults and start doing sweeps. They are no different to Kohya, just that you were probably taught many wrong things. Alpha just multiplies the weights. Leave it at 1 and tune around that. SDXL can have great results even as low as 4, if you can’t get decent results at rank 16 there’s dataset issue. |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
O-J1
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Please read the wiki. Leave values at defaults and start doing sweeps. They are no different to Kohya, just that you were probably taught many wrong things. Alpha just multiplies the weights. Leave it at 1 and tune around that. SDXL can have great results even as low as 4, if you can’t get decent results at rank 16 there’s dataset issue.