Custom Operators Tutorials

Block Floating Point (BFP)

Quantize models using Block Floating Point, a hardware-efficient custom operator supported by Quark.

Quark ONNX Quantization Tutorial For Block Floating Point (BFP)

Microexponents (MX)

Apply Microscaling (MX) formats such as MXFP8 and MXFP4 using Quark’s custom operator support.

Quark ONNX Quantization Tutorial For Image Classification

Custom Operators

Integrate user-defined custom ONNX operators into the Quark quantization pipeline.

Quantizing ONNX Models with Custom Operators Using Quark