3-bit Quantization

A quantization method that represents model weights with a 3-bit bit width. It is commonly used in post-training quantization methods such as GPTQ and AWQ; it achieves a high compression ratio but introduces relatively large perturbations to model behavior.