EpochLex entry

8-Bit Quantization

ayt-bit kwon-tuh-ZAY-shun

Quantization that represents selected model weights or activations with 8-bit numeric precision, commonly to reduce memory use and make inference more efficient than higher-precision representations.