Quants had ruined my Local AI experience. I am hopeful again after using them correctly.

Article automatically generated from technical news.

This is the second time I talk about this here. I started 5 months ago not knowing much. I had just found out that my mac with 32 GB of unified memory could run some decent local models. Everyone recommended 4 bit quants and blabla. Only 1% loss blabla. For months my agentic flows failed badly. Using qwen 27B, 35B, and others. Until I listened to my heart, and to some knowledgeable people, and started using smaller models (like Gemma 4 12B) but with 8Bit quants. No unsloth, no MTP, no d