Intel/llava-gemma-2b
Full fine-tune (fewer layers) vs google/gemma-2b-it. google/gemma-2b-it matches part of Intel/llava-gemma-2b (163 of its 559 tensors; the rest of Intel/llava-gemma-2b has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 18 of 18 layers.
Tensors changed: Norms 36/37, MLP 54/54, Attention 72/72