theairlabcmu/AnyThermal
Full fine-tune vs facebook/dinov2-base. Fine-tune: 98% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 0/5, Attention 96/96, other 24/24, MLP 48/48, Norms 50/50
Full fine-tune vs facebook/dinov2-base. Fine-tune: 98% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 0/5, Attention 96/96, other 24/24, MLP 48/48, Norms 50/50