OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

Tengine

Tengine is a modular C++ deep-learning inference engine for deploying neural-network models on embedded devices. Its architecture separates operator front ends, device back ends, scheduling and execution, and model parsing; the project also provides model-conversion, quantization and benchmarking tools.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Tengine — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Tengine when an embedded application needs a local neural-network inference engine and its documented model, device and toolchain paths match the intended deployment.

Who it is for

  • Developers integrating neural-network inference into embedded or AIoT applications.
  • Teams evaluating CPU, GPU or NPU execution for an embedded deployment.
  • Engineers preparing models for deployment through conversion or uint8/int8 quantization.

Documented uses

  • Run neural-network inference on embedded hardware using a CPU, GPU or NPU back end.
  • Build an embedded image-classification or object-detection application using the included examples.
  • Convert supported model files into Tengine tmfile format.
  • Quantize models to uint8 or int8 and benchmark basic network inference speed.
Before you choose

Requirements & limits

The repository is public and identifies the project as Apache-2.0 licensed. It documents CMake-based compilation and an Ubuntu apt installation and trial guide for x86 and A311D hardware.

  • The project lists CPU, GPU and NPU reference back ends; target-device suitability depends on the relevant back end and deployment path.
  • Tengine uses tmfile for parsed network model parameters, and provides a conversion tool for preparing models for that format.
  • The README describes a prebuilt conversion tool for Ubuntu 18.04, source compilation, and a browser-local WebAssembly conversion option.
  • The quantization tool source supports uint8 and int8; the included benchmark is described as a basic network-speed evaluation tool.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build Tengine’s model conversion tool from source

Compile the conversion utility that converts model files for Tengine deployment.

Before you start

  • A server or PC for the recommended source build.
  • CMake and make, as used by the documented build instructions.

Steps

  1. Create a build directory and enter it: mkdir build && cd build.

  2. Configure the build with conversion-tool support: cmake -DTENGINE_BUILD_CONVERT_TOOL=ON ...

  3. Compile with the command below followed by the system’s nproc command substitution.

    Documented command
    make -j

Expected outcome

A source-built Tengine conversion tool; the README identifies conversion tools as producing models for Tengine, whose serializer parses tmfile model parameters.

Guide limitations

  • The README does not specify supported input model formats or conversion options.
  • The prebuilt converter is described for Ubuntu 18.04; this source-build procedure does not state platform compatibility.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
OAID/Tengine · Official homepage title · retained in its source language
How we review sourcesSuggest a correction