Custom Qwen-27B quant targets 16GB VRAM GPUs

View on Reddit

Reddit user shares IQ4_KS quants for Qwen-27B, optimized for NVIDIA cards with 16GB VRAM. Uses ikawrakow's KS quantization not yet in main llama.cpp.