AuriAetherwiing/G4-31B-Musica-v1
LoRA-style vs google/gemma-4-31B-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 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
12 derivatives checked
- bartowski/AuriAetherwiing_G4-31B-Musica-v1-GGUF: Same weights, converted (fewer layers)
- mradermacher/G4-31B-Musica-v1-i1-GGUF: Quantization (fewer layers)
- mradermacher/G4-31B-Musica-v1-GGUF: Quantization (fewer layers)
- Ateron/Gemma-4-MoonGem-31B: Full fine-tune
- Blazed-Forge/Gemma-4-Gemsicle-31B: LoRA-style
- Blazed-Forge/Gemma-4-Giftige-Blume-31B-v1: LoRA-style
- Nimbz/Gemma-4-Gembrain-X-Core-31B: Full fine-tune
- Nimbz/Gemma-4-Gembrain-31B: LoRA-style
- Nimbz/Gemma-4-Gembrain-X-31B: Full fine-tune
- Nimbz/Gemma-4-Giftige-Blume-31B-v2: LoRA-style
- Nimbz/Prosopon-31B: Full fine-tune
- Nimbz/Versipellis-31B: LoRA-style