TensorFlow Lite for Microcontrollers is designed to run deep learning models on bare-metal systems, DSPs, and microcontrollers with limited memory (tens to hundreds of kilobytes). It runs models converted via LiteRT and TensorFlow.

• Tier-1 Architectures (Tested in CI): Arm Cortex-M (CMSIS-NN / Corstone), Cadence Xtensa (HiFi 3/4/5, Vision P6), RISC-V, Qualcomm Hexagon, and x86/Windows.
• Community Platforms & SDKs: Arduino, Coral Dev Board Micro, Espressif ESP32, Renesas, Silicon Labs, and Ingenic MIPS.

• Core Concepts:
  • Memory Management
  • Error Handling & Defensive Programming
  • Logging
  • Continuous Integration
• Porting & Hardware Optimization:
  • Optimized Kernel Implementations
  • New Platform Support
  • Arm IP Support
  • Software Emulation (Renode) and QEMU
• Tooling, Compression & Python:
  • Model Compression Guide & Tutorial
  • Python Interpreter & Dev Guide
  • Layer-by-Layer Debugging Tool
  • Benchmarks and Profiling

• Questions & Issues: Please file a GitHub Issue.
• Contributing: Review our Contributing Guide.
• Security: Report vulnerabilities according to our Security Policy.