llama.cpp

Restore intended k-quants quantization mixes for MoE models

#4872

Merged