amalia-llm/AMALIA-VL-DPO

Full fine-tune vs amalia-llm/AMALIA-VL-SFT. Fine-tune: 90% of the weights changed, across 42 of 42 layers.

Tensors changed: other 5/5, Embeddings 0/4, Norms 29/195, MLP 126/234, Attention 168/384, lm_head 1/1

1 derivatives checked

• orbcom-pedroferreira/AMALIA-VL-DPO-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse