tensorflow-onnx

Why we included this project

For teams that train in TensorFlow or Keras but serve on ONNX-based runtimes, this tool handles the format gap. It converts saved models, checkpoints, graph definitions, TFLite, and (experimentally) TensorFlow.js into a standard ONNX graph that ONNX Runtime and similar runtimes can execute. You can run conversions from the command line or through the Python API, and you can set the output opset to match your target runtime. TensorFlow has more operators than ONNX, so not every exotic op maps cleanly; check the supported-ops list and troubleshooting guide before relying on a conversion.