pytorch/QNNPACK

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

1551 stars · 223 forks observed · 2026-08-28

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

QNNPACK is a mobile-optimized C library of high-performance kernels for 8-bit quantized neural network operators such as convolution, pooling, and fully connected layers. It serves as a low-level performance primitive integrated into PyTorch (via Caffe2) rather than being used directly by researchers.

Use cases

• run quantized neural network inference on mobile phones
• speed up depthwise convolutions in MobileNet-style models on ARM devices
• cross-compile quantized operator kernels for Android and iOS
• integrate fast 8-bit inference primitives into a deep learning framework
• perform real-time image classification on low-power mobile hardware

When to choose

• you need fast quantized 8-bit inference on ARM mobile CPUs
• you are building or optimizing a deep learning runtime for Android or iOS
• you need low-level NEON-optimized kernels for convolutions and pooling

When to avoid

• you want a high-level framework or model training tool
• you target x86 servers or GPU inference
• you need a maintained library with recent updates and broad operator coverage

Facets

library · maturity maintenance

machine-learning deep-learning llm-inference machine-learning mobile-development computer-vision deep-learning cross-platform cpp quantization inference-kernels arm-neon mobile-ai 8-bit-operators android ios

Member repositories

For agents

markdown · JSON · MCP: product_card(name="pytorch/QNNPACK")