Efficient-Large-Model/Fast_dVLM_3B
Partial fine-tune vs Qwen/Qwen2.5-VL-3B-Instruct. Partial change: 82% of the weights changed (attention, embeddings, mlp, norms), in 36 of 36 layers.
Tensors changed: Embeddings 1/2, Norms 73/138, MLP 108/304, Attention 252/380