OpenGVLab/InternVL3_5-30B-A3B-MPO

Full fine-tune vs OpenGVLab/InternVL3_5-30B-A3B-Instruct. Fine-tune: 94% of the weights changed, across 48 of 48 layers.

Tensors changed: Norms 199/289, MLP 2364/2369, Attention 288/288, lm_head 1/1, Embeddings 4/5, other 0/48

2 derivatives checked

• OpenGVLab/InternVL3_5-30B-A3B: Partial fine-tune
• OpenGVLab/InternVL3_5-30B-A3B-HF: Full fine-tune

On the Hub · Downloads on Model Pulse