# OAID/Tengine

Tengine is a lite, high performance, modular inference engine for embedded device

Repository: https://github.com/OAID/Tengine
Canonical: https://ross.abutalabs.com/products/oaid-tengine
Language: C++
License: Apache-2.0
License Family: permissive
Topics: arm, machine-learning, artificial-intelligence, cnn, tensorflow, pytorch, onnx, x86-64, mips, cuda, tensorrt, acl, npu, riscv, supperedge, container, nvdla
Last push: 2025-03-06T07:12:55+00:00

## Health v2 (maintenance only)
Score: 27/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 10, release rhythm 8, longevity 100
- inputs: {"age_days": 3169, "days_push": 545, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4531, forks 981 (observed 2026-08-28T04:08:52.991870+00:00)

## What it is
Tengine is a lightweight, high-performance, modular deep learning inference engine developed by OPEN AI LAB for embedded and edge devices. It features a decoupled frontend/backend architecture supporting CPU, GPU, and NPU heterogeneous compute, with model conversion and quantization tools for frameworks like TensorFlow, PyTorch, and ONNX.

## Use cases
- deploy cnn models on embedded arm devices
- run inference on npu hardware like verisilicon tim-vx
- convert onnx or tensorflow models to a lightweight runtime format
- quantize models to int8 for faster edge inference
- benchmark neural network speed on embedded boards
- deploy ai models in edge computing containers with superedge
- run deep learning inference on riscv or mips devices

## When to choose
- you need efficient on-device inference on resource-constrained embedded hardware
- you want to target heterogeneous backends including cpu, gpu, and npu from one framework
- you need a small-footprint c/c++ runtime for aiot applications
- you require model conversion and quantization tooling for edge deployment

## When to avoid
- you need training or fine-tuning capabilities rather than inference
- you primarily target cloud or server environments with abundant resources
- you depend on a broad ecosystem of prebuilt ops and community extensions like onnxruntime or tflite
- you need first-class support for large language models or transformer architectures

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, llm-inference, gpu-computing, embedded, benchmarking, cli
- domain: machine-learning, deep-learning, embedded-systems, iot, artificial-intelligence, computer-vision, cross-platform
- platform: embedded, cpp, cross-platform
- tags: inference-engine, npu, aiot, edge-ai, onnx, model-conversion, quantization, tim-vx, riscv, tensorrt, linux, arm, gpu, android

## Member repositories
- OAID/Tengine (main) score 27

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:52.991870+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 2026-08-29T18:20:12.868967+00:00, confidence not recorded.
  - readme: https://github.com/OAID/Tengine (fetched 2026-08-28T04:08:52.991870+00:00, sha 7a5c3c7528de)
- Data as of 2026-08-30T08:39:29.467469+00:00.