AbstractPhil's picture
Update README.md
aef4253 verified
metadata
license: mit

1/32/2025 Lailah Isn't ready yet: step_417054.pt

Baseline ablation tests ran with simple 10 image dataset.

The model requires a large amount more pretraining at a higher learn rate to establish a firmer grasp of anything useful.

She's not ready for useful loras yet.

This model potentially requires additional single blocks, likely to match 2:1 single to double ratio; but I will not alter the model until I'm certain.