Module scheme 

Quantization Schemes

This module exposes a suite of post-training quantization (PTQ) strategies:

Module	Scheme	Primary use
minmax	Min-Max calibration (INT4/INT8)	General PTQ
nf4	NormalFloat4 (QLoRA)	4-bit weights
fp8	FP8 E4M3 / E5M2 (Hopper / Blackwell)	Training & inference
gptq	GPTQ Hessian-guided quantization	LLM weights
smooth_quant	SmoothQuant activation–weight migration	LLM activations

Re-exports

pub use fp8::Fp8Codec;

pub use fp8::Fp8Format;

pub use gptq::GptqConfig;

pub use gptq::GptqOutput;

pub use gptq::GptqQuantizer;

pub use minmax::MinMaxQuantizer;

pub use minmax::QuantGranularity;

pub use minmax::QuantParams;

pub use minmax::QuantScheme;

pub use nf4::NF4_LUT;

pub use nf4::Nf4Quantizer;

pub use smooth_quant::SmoothQuantConfig;

pub use smooth_quant::SmoothQuantMigrator;

Modules

fp8

FP8 Floating-Point Quantization

gptq

GPTQ — Gradient-Free Post-Training Quantization

minmax

MinMax Quantizer

nf4

NF4 — NormalFloat4 Quantization

smooth_quant

SmoothQuant — Activation–Weight Quantization Migration