Ateron/Gemma-4-Writers-31B-V2
LoRA-style vs ConicCat/Gemma4-Writer-31B-D. 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 60 of 60 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, Norms 0/523, other 0/62, MLP 180/261, Attention 230/338
4 derivatives checked
- mradermacher/Gemma-4-Writers-31B-V2-i1-GGUF: Quantization (fewer layers)
- mradermacher/Gemma-4-Writers-31B-V2-GGUF: Quantization (fewer layers)
- ailexleon/Gemma-4-Writers-31B-V2-mlx-8Bit: Quantization
- ailexleon/Gemma-4-Writers-31B-V2-mlx-4Bit: Quantization