Llama.cpp quantization is broken
Main reason is, that qunatization quality directly affects models performance and stability and this results in real usefullness. Even though GRM-2.6-Plus is in benchmarks better than qwen3.6 27b model from which it derives, it gives worse result…