The convention is weird but it's pretty standard in the industry across all models, particularly GGUF. 671B parameters, quantized to 4 bits. The K_M terminology

The convention is weird but it's pretty standard in the industry across all models, particularly GGUF. 671B parameters, quantized to 4 bits. The K_M terminology I believe is more specific to GGUF and describes the specific quantization strategy.