OpenGVLab/InternVL3_5-14B-Instruct

Full fine-tune vs OpenGVLab/InternVL3_5-14B-Pretrained. Fine-tune: 100% of the weights changed, across 40 of 40 layers.

Tensors changed: Embeddings 5/5, Norms 257/257, MLP 222/222, Attention 256/256, other 48/48, lm_head 1/1

1 derivatives checked