llm-jp/waon-siglip2-base-patch16-256
Full fine-tune vs google/siglip2-base-patch16-256. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: other 5/5, Embeddings 5/5, Norms 102/102, MLP 100/100, Attention 196/196
Full fine-tune vs google/siglip2-base-patch16-256. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: other 5/5, Embeddings 5/5, Norms 102/102, MLP 100/100, Attention 196/196