NbAiLab/nb-sau-13b-4k-step100k
Partial fine-tune vs meta-llama/Llama-2-13b-hf. Partial change: 33% of the weights changed (attention, norms), in 40 of 40 layers.
Tensors changed: Norms 81/81, MLP 0/120, Attention 160/160
Partial fine-tune vs meta-llama/Llama-2-13b-hf. Partial change: 33% of the weights changed (attention, norms), in 40 of 40 layers.
Tensors changed: Norms 81/81, MLP 0/120, Attention 160/160