JHL3/KLOCR
Full fine-tune vs team-lucid/trocr-small-korean. Fine-tune: 99% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 8/9, Attention 192/192, Norms 86/86, MLP 24/24, other 49/51
Full fine-tune vs team-lucid/trocr-small-korean. Fine-tune: 99% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 8/9, Attention 192/192, Norms 86/86, MLP 24/24, other 49/51