Hcompany/Holotron-12B

Full fine-tune vs nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16. Fine-tune: 95% of the weights changed, across 62 of 62 layers.

Tensors changed: Embeddings 2/3, other 131/197, Norms 165/219, MLP 185/187, Attention 152/152, lm_head 1/1

On the Hub · Downloads on Model Pulse