Q4 Quantization

A technique that compresses model weights into approximately 4-bit integer representations. It can significantly reduce memory usage and accelerate inference, and is often combined with mixed-precision strategies.