Axon NPU compiler

The following pages describe the TinyML sample models provided with the Axon TFLite compiler. They cover instructions on how the compiler is used to compile and run machine learning models on the Axon NPU. The TinyML models serve as reference implementations and show how to prepare data, compile models using the local compiler, and validate inference on Axon. Each model includes an overview and a dedicated subpage with model‑specific details and usage instructions.

• Supported operators
  • Model structure
  • Memory organization and reshape
  • Operators
  • Model design recommendations
• Axon DSP intrinsics
  • Fixed-point arithmetic
  • Synchronous execution
  • Performance benefits compared to the CPU
  • Developing with Axon intrinsics
  • Intrinsic calling conventions
  • Intrinsic listing
• Axon NPU TFLITE compiler
  • Overview
  • Directory layout
  • Setting up the executor
  • Configuring input parameters
  • Sample TinyML models
  • Running the Compiler
  • Using Docker (Optional)
  • Alternative to Docker: Podman
  • Troubleshooting
  • Verifying model support (scanner script)
  • Verifying model on an Axon NPU-enabled device
• MLPerf™ Tiny Models
  • Overview
  • Evaluation datasets
  • Model setup and training
  • Setting up the Python environment
  • Available models
• Axon NPU changelog
  • Changelog