OpenGVLab/InternVL3_5-241B-A28B-MPO

Full fine-tune vs OpenGVLab/InternVL3_5-241B-A28B-Instruct. Fine-tune: 96% of the weights changed, across 94 of 94 layers.

Tensors changed: Norms 328/557, MLP 1821/1836, Attention 511/511, lm_head 1/1, Embeddings 3/5, other 86/90

2 derivatives checked