HuggingFaceTB/SmolVLM-Synthetic
Full fine-tune (fewer layers) vs HuggingFaceTB/SmolLM2-1.7B-Instruct. HuggingFaceTB/SmolLM2-1.7B-Instruct matches part of HuggingFaceTB/SmolVLM-Synthetic (217 of its 657 tensors; the rest of HuggingFaceTB/SmolVLM-Synthetic has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 24 of 24 layers.
Tensors changed: Norms 49/49, MLP 72/72, Attention 96/96