JoPmt/Hermetic-Llama-Ties
Partial fine-tune vs BEE-spoke-data/smol_llama-220M-GQA. Partial change: 85% of the weights changed (attention, lm_head, mlp), in 10 of 10 layers.
Tensors changed: lm_head 1/1, Embeddings 0/1, Norms 0/21, MLP 30/30, Attention 40/40
1 derivatives checked
- mradermacher/Hermetic-Llama-Ties-GGUF: Quantization