SteelStorage/Aura-llama
Partial fine-tune vs NousResearch/Meta-Llama-3-8B-Instruct. Partial change: 54% of the weights changed (attention, mlp, norms), in 20 of 32 layers.
Tensors changed: Embeddings 0/1, Norms 40/65, MLP 60/96, Attention 80/128, lm_head 0/1
2 derivatives checked
- mradermacher/Aura-llama-i1-GGUF: Quantization
- mradermacher/Aura-llama-GGUF: Same weights, converted