soob3123/amoral-gemma3-4B-v2-qat
LoRA-style vs google/gemma-3-4b-it-qat-q4_0-unquantized. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 34 of 34 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/316, MLP 0/210, Attention 136/352, other 0/1
3 derivatives checked
- mradermacher/amoral-gemma3-4B-v2-qat-i1-GGUF: Quantization (fewer layers)
- mradermacher/amoral-gemma3-4B-v2-qat-GGUF: Same weights, converted (fewer layers)
- soob3123/amoral-gemma3-4B-v2-qat-Q4_0-GGUF: Quantization (fewer layers)