DS-Archive/L3.3-70B-Magnum-v5-SFT-Alpha

Full fine-tune vs meta-llama/Llama-3.3-70B-Instruct. Fine-tune: 99% of the weights changed, across 80 of 80 layers.

Tensors changed: Embeddings 0/1, Norms 0/161, MLP 240/240, Attention 320/320, lm_head 1/1

3 derivatives checked