int8 quantization

A quantization technique that compresses model weights from 16-bit or 32-bit floating point into 8-bit integer storage, significantly reducing memory footprint. It is often combined with quantization-aware training (QAT) to compensate for accuracy loss.