ai-forever/kandinsky-video-motion-predictor
Full fine-tune vs OpenGVLab/VideoMAEv2-Base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 60/60, MLP 48/48, Norms 50/50, Embeddings 2/2
Full fine-tune vs OpenGVLab/VideoMAEv2-Base. Fine-tune: 100% of the weights changed, across 12 of 12 layers.
Tensors changed: Attention 60/60, MLP 48/48, Norms 50/50, Embeddings 2/2