Q4_K_M

A mixed quantization scheme in the GGUF format that uses 4-bit quantization. Through k-quant block techniques it retains higher precision in key layers, balancing model size against output quality.