wuzhi-hao/EgoSim
Full fine-tune vs alibaba-pai/Wan2.1-Fun-14B-InP. Fine-tune: 94% of the weights changed, across 40 of 40 layers.
Tensors changed: Attention 760/800, Norms 54/280, MLP 160/160, other 52/53, Embeddings 9/9
Full fine-tune vs alibaba-pai/Wan2.1-Fun-14B-InP. Fine-tune: 94% of the weights changed, across 40 of 40 layers.
Tensors changed: Attention 760/800, Norms 54/280, MLP 160/160, other 52/53, Embeddings 9/9