worstchan/EAT-base_epoch30_finetune_AS2M
Full fine-tune vs worstchan/EAT-base_epoch30_pretrain. Fine-tune: 95% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 48/48, MLP 48/48, Norms 50/50, other 3/4
Full fine-tune vs worstchan/EAT-base_epoch30_pretrain. Fine-tune: 95% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 48/48, MLP 48/48, Norms 50/50, other 3/4