Archatext/hatformer-arams28k
Full fine-tune vs microsoft/trocr-base-stage1. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 5/5, Attention 252/252, Norms 122/122, MLP 48/48, other 48/48
Full fine-tune vs microsoft/trocr-base-stage1. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 5/5, Attention 252/252, Norms 122/122, MLP 48/48, other 48/48