nisten/BigCodeLlama-92b
Partial fine-tune vs codellama/CodeLlama-70b-Instruct-hf. Partial change: 14% of the weights changed (attention, lm_head, mlp, norms), in 11 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 23/161, MLP 33/240, Attention 44/320, lm_head 1/1
1 derivatives checked
- mradermacher/BigCodeLlama-92b-i1-GGUF: Quantization