ReadyArt/Dark-Scarlett-v1.0-26B-A4B
LoRA-style vs google/gemma-4-26B-A4B-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 30 of 30 layers; norms, biases and embeddings are untouched.
Tensors changed: Embeddings 0/4, MLP 90/231, Norms 0/433, other 0/122, Attention 115/223
6 derivatives checked
- ReadyArt/Dark-Scarlett-v1.0-26B-A4B-GGUF: Quantization
- mradermacher/Dark-Scarlett-v1.0-26B-A4B-i1-GGUF: Quantization
- mradermacher/Dark-Scarlett-v1.0-26B-A4B-GGUF: Quantization
- Wiself/Dark-Scarlett-v1.0-26B-A4B-Voice: Same weights, repacked (fewer layers)
- Naphula/Promethean-Dawn-26B-A4B: Partial fine-tune (fewer layers)
- Vortex5/G4-Midnight-Macaw-26B-A4B: Full fine-tune