OpenPPL/ppq

Health v2 · maintenance only

Adoption not part of the score

1816 stars · 285 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

PPQ (PPL Quantization Tool) is an extensible, high-performance offline neural network quantization framework written in Python. It parses and quantizes complex model graphs (ONNX, PyTorch, Caffe), supports FP8 and QAT, and exports to inference engines like TensorRT, OpenPPL, OpenVINO, ncnn, and SNPE.

Use cases

• quantize a neural network model for edge deployment
• convert an onnx model to int8 for tensorrt
• simulate quantized inference on different hardware platforms
• perform quantization-aware training on a model
• calibrate quantization parameters for a custom accelerator
• export a quantized model to ncnn or openvino

When to choose

• you need fine-grained control over per-operator quantization settings and custom quantization passes
• you target multiple inference backends like TensorRT, OpenPPL, ncnn, or SNPE
• you need FP8 or QAT support alongside standard PTQ calibration
• you want to experiment with quantization research on custom hardware

When to avoid

• you just need simple one-click post-training quantization with minimal configuration
• your workflow is TensorFlow-only, since PPQ focuses on ONNX, PyTorch, and Caffe
• you need a maintained tool with frequent releases, as the latest release dates to early 2024

Facets

library · maturity maintenance

machine-learning deep-learning compiler benchmarking parser deep-learning machine-learning gpu-computing developer-tools python windows quantization model-optimization onnx pytorch inference-deployment qat fp8 neural-network-acceleration linux cuda gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="OpenPPL/ppq")