kinfer 0.3.3

Tool to make it easier to run a model on a real robot

kinfer

This package is designed to support exporting and running inference on PyTorch models.

Installation

pip install kinfer

ONNX Runtime

You can install the latest version of ONNX Runtime on Mac with:

brew install onnxruntime

You may need to add the binary to your DYLD_LIBRARY_PATH:

$ brew ls onnxruntime
/opt/homebrew/Cellar/onnxruntime/1.20.1/include/onnxruntime/ (11 files)
/opt/homebrew/Cellar/onnxruntime/1.20.1/lib/libonnxruntime.1.20.1.dylib  # <-- This is the binary
/opt/homebrew/Cellar/onnxruntime/1.20.1/lib/cmake/ (4 files)
/opt/homebrew/Cellar/onnxruntime/1.20.1/lib/pkgconfig/libonnxruntime.pc
/opt/homebrew/Cellar/onnxruntime/1.20.1/lib/libonnxruntime.dylib
/opt/homebrew/Cellar/onnxruntime/1.20.1/sbom.spdx.json
$ export DYLD_LIBRARY_PATH=/opt/homebrew/Cellar/onnxruntime/1.20.1/lib:$DYLD_LIBRARY_PATH

Considerations for Exporting PyTorch Models

Don't use common names for the inputs to your forward pass. E.g. input, output, state, state_tensor, buffer, etc.

This is because ONNX has internal names for the model and if there's a conflict, the inputs will have a .1, .2, etc. suffix which makes it really hard to figure out what value_name to pass into your kinfer io values.