Python tests and samples

The tests/python_tests/ directory contains a pytest suite that exercises the EP through the public ONNX Runtime Python API.

Setup

Set up a virtual environment from the repo root:

python -m venv .venv-pytests
.\.venv-pytests\Scripts\Activate.ps1
pip install -r tests\python_tests\requirements.txt
pip install path\to\onnxruntime_ep_nv_tensorrt_rtx-*.whl

Use the onnxruntime version from the release notes for your EP release.

Run

Run pytest from inside tests\python_tests so test artifacts land next to the tests rather than in the repo root:

cd tests\python_tests
python -m pytest -v

Coverage

Test file	What it covers
`test_registration.py`	EP library registration, device discovery, unregister and re-register
`test_sessions.py`	Session creation with explicit EP devices and provider options
`test_ep_context.py`	EPContext creation and reload via `ep.context_*` session options
`test_compile_model.py`	Compile API file and bytes coverage (requires `onnxruntime.ModelCompiler`)
`test_proto_preprocessing.py`	Proto-preprocessing passes exercised indirectly through compile

Notes

Tests require an NVIDIA RTX GPU to run.
The compile tests (test_compile_model.py) require onnxruntime 1.27+ which includes onnxruntime.ModelCompiler. No special build needed — pip install onnxruntime>=1.27 is sufficient.
Test artifacts (ONNX models, EPContext files, cache directories) are written to the current working directory and are not auto-cleaned between runs.