# google/XNNPACK

High-efficiency floating-point neural network inference operators for mobile, server, and Web

Repository: https://github.com/google/XNNPACK
Canonical: https://ross.abutalabs.com/products/xnnpack
Language: C
License: NOASSERTION
License Family: other
Topics: neural-networks, inference, inference-optimization, simd, cpu, multithreading, matrix-multiplication, convolutional-neural-networks, convolutional-neural-network, neural-network, mobile-inference
Last push: 2026-08-26T18:42:51+00:00

## Health v2 (maintenance only)
Score: 77/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 100
- inputs: {"age_days": 2546, "days_push": 7, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2435, forks 547 (observed 2026-08-28T04:06:51.388155+00:00)

## What it is
XNNPACK is a highly optimized C library of floating-point neural network inference operators for ARM, x86, WebAssembly, and RISC-V platforms. It serves as a low-level performance primitive layer used by high-level frameworks like TensorFlow Lite, PyTorch, ONNX Runtime, and MediaPipe rather than being used directly by practitioners.

## Use cases
- accelerate neural network inference on mobile devices
- optimize convolution and matrix multiplication kernels for ARM and x86
- run neural network operators in the browser via WebAssembly SIMD
- speed up TensorFlow Lite or PyTorch mobile inference
- implement efficient SIMD-optimized deep learning primitives
- benchmark single-threaded inference performance on phones

## When to choose
- you are building or optimizing an ML inference framework and need fast low-level kernels
- you need cross-architecture support including ARM, x86, RISC-V, and WebAssembly
- you need fine-grained control over operator performance, threading, and memory layouts like NHWC with channel slicing

## When to avoid
- you are a deep learning practitioner looking for a high-level training or inference API
- you need GPU acceleration rather than CPU-optimized operators
- you want a ready-to-use end-user application rather than a library of primitives

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, benchmarking, concurrency
- domain: machine-learning, deep-learning, mobile-development, web-development, performance, embedded-systems
- platform: windows, wasm, cross-platform, cpp
- tags: neural-network-inference, simd-optimization, kernel-library, arm-neon, risc-v, inference-operators, low-level-primitives, android, ios, macos, linux

## Member repositories
- google/XNNPACK (main) score 77

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:51.388155+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:31:02.393342+00:00, confidence not recorded.
  - readme: https://github.com/google/XNNPACK (fetched 2026-08-28T04:06:51.388155+00:00, sha bd6e92442bcf)
- Data as of 2026-08-30T08:39:29.467469+00:00.