LLM Binarization — Episode One: DBF aka Addition Is (Almost) All You Need
A runnable DBF/SIGN1 model: near-Q3_K quality at a slightly smaller-than-Q2_K budget and near-Q2_K Vulkan speed.
A runnable DBF/SIGN1 model: near-Q3_K quality at a slightly smaller-than-Q2_K budget and near-Q2_K Vulkan speed.