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.
👋 Open to Work
Nicolo' D'Evangelista
Ninnix96
AI & ML interests
None yet
Recent Activity
upvoted a collection 6 days ago
💧 LFM3 updated a model 11 days ago
Ninnix96/Qwen3.6-35B-A3B-gguf updated a collection 11 days ago
Antirez's MoE recipeOrganizations
None yet