nvidia/VideoITG-8B
Full fine-tune (fewer layers) vs Qwen/Qwen2-7B-Instruct. Qwen/Qwen2-7B-Instruct matches part of nvidia/VideoITG-8B (338 of its 792 tensors; the rest of nvidia/VideoITG-8B has no counterpart). On the shared part: Fine-tune: 92% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 57/57, MLP 84/84, Attention 196/196