GJLさんがSDXLモデルの高精度FP8量子化の方法を公開されました。 HSWQ法という手法です。 内容については何回かにわたってGJLさんのNote記事に詳細に記載されていますので、興味がある方は是非。 ざっくりというと、オーダーメイド型のFP8変換をすることで ...

This workflow automates the process of converting a standard float32 neural network model (in ncnn format) into an int8 quantized model. Quantization replaces 32-bit floating-point weights and ...

Model quantization converts the high-precision floating-point weights in a neural network (32-bit or 16-bit) into compact lower-precision representations (8-bit, 4-bit, or even 2-bit integers). This ...

Reducing the precision of model weights can make deep neural networks run faster in less GPU memory, while preserving model accuracy. If ever there were a salient example of a counter-intuitive ...