EpochLex entry

4-Bit Quantization

four-bit kwon-tuh-ZAY-shun

A model-compression approach that represents model weights or other numerical values using 4-bit precision, substantially reducing memory requirements while potentially changing accuracy or performance.