Antirez recipe for MoE quantization: Q2 uses IQ2_XXS gate/up + Q2_K down; Q4 uses Q4_K gate/up + MXFP4 down. Everything else almost unchanged.
-
Ninnix96/Qwen3.6-35B-A3B-gguf
Text Generation • 35B • Updated • 1.92k • 2 -
Ninnix96/Mellum2.1-12B-A2.5B-Thinking-gguf
Text Generation • 12B • Updated • 622 -
Ninnix96/gemma-4-26B-A4B-gguf
Text Generation • 25B • Updated • 632 -
Ninnix96/Ling-3.0-tiny-gguf
Text Generation • 8B • Updated • 835 • 1