anaspro/Shako-iraqi-4B-it
LoRA-style vs unsloth/gemma-3n-E4B-it. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 35 of 35 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 140/363, Embeddings 0/25, Norms 0/489, other 0/687, MLP 105/112
3 derivatives checked
- mradermacher/Shako-iraqi-4B-it-GGUF: Same weights, converted (fewer layers)
- mradermacher/Shako-iraqi-4B-it-i1-GGUF: Quantization (fewer layers)
- mradermacher/Shako-4B-it-GGUF: LoRA-style (fewer layers)