egeorcun/lucida
Full fine-tune vs ZhengPeng7/BiRefNet_HR. Fine-tune: 99% of the weights changed, across 4 of 4 layers.
Tensors changed: Attention 126/150, other 394/394, MLP 96/96, Norms 110/110, Embeddings 4/4
Full fine-tune vs ZhengPeng7/BiRefNet_HR. Fine-tune: 99% of the weights changed, across 4 of 4 layers.
Tensors changed: Attention 126/150, other 394/394, MLP 96/96, Norms 110/110, Embeddings 4/4