ai-forever/kandinsky-videomae-large-camera-motion
Full fine-tune (fewer layers) vs MCG-NJU/videomae-large. ai-forever/kandinsky-videomae-large-camera-motion matches part of MCG-NJU/videomae-large (364 of its 550 tensors; the rest of MCG-NJU/videomae-large has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Embeddings 2/2, Attention 168/168, other 96/96, Norms 98/98