cnmoro/mini-image-captioning
Full fine-tune (fewer layers) vs WinKawaks/vit-small-patch16-224. WinKawaks/vit-small-patch16-224 matches part of cnmoro/mini-image-captioning (198 of its 316 tensors; the rest of cnmoro/mini-image-captioning has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Embeddings 4/4, Attention 96/96, other 48/48, Norms 50/50