Quantization is an umbrella term that covers a lot of different techniques, but it basically involves the process of converting continuous infinite input values from a large set to discrete finite output values in a smaller set. The process reduces the precision of numerical representations in a model. The goal of quantization is to reduce the number of bits needed to represent information. This makes the model more efficient in terms of memory usage, storage, and computational resources while preserving its performance to a reasonable extent, resulting in higher performance.

While quantization offers memory and performance advantages, it can introduce challenges, including the potential drop in model accuracy due to reduced precision. Careful optimization and fine-tuning are essential to mitigate these challenges.

References: https://www.qualcomm.com/news/onq/2019/03/heres-why-quantization-matters-ai