open-machine/Llama-3.2-1B-FlashNorm-test
Partial fine-tune vs meta-llama/Llama-3.2-1B. Partial change: 52% of the weights changed (attention, mlp, norms), in 16 of 16 layers.
Tensors changed: Embeddings 0/1, Norms 32/33, MLP 32/48, Attention 48/64
2 derivatives checked
- mradermacher/Llama-3.2-1B-FlashNorm-i1-GGUF: Quantization
- mradermacher/Llama-3.2-1B-FlashNorm-GGUF: LoRA-style