Ray2333/GRM-llama3.2-3B-sftreg

Full fine-tune vs meta-llama/Llama-3.2-3B-Instruct. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: Norms 34/57, MLP 84/84, Attention 112/112

1 derivatives checked

• Ray2333/GRM-Llama3.2-3B-rewardmodel-ft: Partial fine-tune

On the Hub · Downloads on Model Pulse