onnx-quantize

ONNX Quantization Framework built on top of

• ONNX IR
• ONNXScript

⚠️ This project is under active development.

---

📦 Installation

Install directly from PyPI:

pip install onnx-quantize

🚀 Quick Start

Here’s a minimal example to quantize an ONNX model:

from onnx_quantize import QConfig, QuantType, quantize 
import onnx

# Load your model
model = onnx.load("your_model.onnx")

# Define quantization configuration
qconfig = QConfig(
    is_static=False,
    activations_dtype=QuantType.QInt8,
    activations_symmetric=False,
    weights_dtype=QuantType.QInt8,
    weights_symmetric=True,
    weights_per_channel=False,
)

# Quantize the model
qmodel = quantize(model, qconfig)

# Save the quantized model
onnx.save(qmodel, "qmodel.onnx")

🧩 Features (planned)

The goal is to have all of what Neural compressor have but using ONNXScript and ONNX IR.