ValiantLabs/gemma-4-12B-it-Esper4
LoRA-style vs google/gemma-4-12B-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 48 of 48 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/12, Norms 0/289, other 0/48, MLP 144/144, Attention 184/184
8 derivatives checked
- bartowski/ValiantLabs_gemma-4-12B-it-Esper4-GGUF: Same weights, converted
- mradermacher/gemma-4-12B-it-Esper4-i1-GGUF: Quantization
- mradermacher/gemma-4-12B-it-Esper4-GGUF: Quantization
- OliviaRossi/gemma-4-12B-EsperGrug: LoRA-style
- EryriLabs/Caelum-G4-38B-A12.5B: LoRA-style
- sequelbox/gemma-4-12B-it-Esper4FableComposer: Full fine-tune
- sequelbox/gemma-4-12B-it-Esper4Grug: LoRA-style
- sequelbox/gemma-4-12B-it-Esper4Tachibana: LoRA-style