OpenGVLab/InternVL3_5-4B-MPO

Full fine-tune vs OpenGVLab/InternVL3_5-4B-Instruct. Fine-tune: 92% of the weights changed, across 36 of 36 layers.

Tensors changed: Embeddings 4/5, Norms 144/241, MLP 205/210, Attention 240/240, other 0/48, lm_head 1/1

4 derivatives checked

• OpenGVLab/InternVL3_5-4B: Partial fine-tune
• OpenGVLab/InternVL3_5-4B-HF: Full fine-tune
• OpenGVLab/InternVL3_5-2B-HF: Different architecture
• bairongz/QianfanOCR: Full fine-tune

On the Hub · Downloads on Model Pulse