MCG-NJU/UniAVGen
Full fine-tune (fewer layers) vs Wan-AI/Wan2.2-TI2V-5B. Wan-AI/Wan2.2-TI2V-5B matches part of MCG-NJU/UniAVGen (821 of its 2044 tensors; the rest of MCG-NJU/UniAVGen has no counterpart). On the shared part: Fine-tune: 99% of the weights changed, across 30 of 30 layers.
Tensors changed: Attention 480/480, Norms 180/180, MLP 120/120, other 33/35, Embeddings 6/6