tensorflow/tflite-micro

Infrastructure to enable deployment of ML models to low-power resource-constrained embedded targets (including microcontrollers and digital signal processors).

What it solves

It enables the execution of machine learning models on hardware with extremely limited memory, such as microcontrollers and Digital Signal Processors (DSPs).

How it works

It is a specialized port of TensorFlow Lite designed specifically for resource-constrained environments, allowing models to run directly on the device hardware.

Who it’s for

Embedded systems developers and engineers working with microcontrollers (e.g., ARM Cortex-M, RISC-V, Xtensa) and DSPs.

Highlights

  • Support for a wide range of platforms including Arduino, Coral Dev Board Micro, and boards from Espressif, Renesas, and Texas Instruments.
  • Specialized documentation for memory management, profiling, and error handling in constrained environments.
  • Support for optimized kernel implementations and model compression.
  • Integration with software emulation tools like Renode and QEMU.

Related

  • Project
    tensorflow/tensorflow

    An end-to-end open source platform for machine learning that provides a flexible ecosystem of tools and libraries for building and deploying ML-powered applications.

  • Project
    mattn/go-tflite

    A Go language binding for TensorFlow Lite that allows developers to run lightweight machine learning models and perform inference within Go applications.

  • Project
    espressif/esp-tflite-micro

    A component for the ESP-IDF platform that enables running TensorFlow Lite Micro on Espressif chipsets with hardware-optimized kernels for faster inference.

  • Project
    tensorflow/tfjs

    A hardware-accelerated JavaScript library for training and deploying machine learning models in the browser and Node.js.

  • Project
    vitoplantamura/OnnxStream

    A memory-optimized inference library that enables large AI models like Stable Diffusion and LLMs to run on ultra-low-RAM devices by streaming weights and using attention slicing.