facebook/sapiens2-pointmap-5b
Full fine-tune vs facebook/sapiens2-pretrain-5b. Fine-tune: 100% of the weights changed, across 56 of 56 layers.
Tensors changed: Attention 504/504, Norms 225/225, MLP 224/224, Embeddings 2/3
Full fine-tune vs facebook/sapiens2-pretrain-5b. Fine-tune: 100% of the weights changed, across 56 of 56 layers.
Tensors changed: Attention 504/504, Norms 225/225, MLP 224/224, Embeddings 2/3