AINode requires Python 3.11 and uses Torch 2.14.1, setuptools 84.2.1, and PyInstaller 6.22.3. Linux selects the official Torch CUDA 12.6 wheels from https://download.pytorch.org/whl/cu126 to retain Volta GPU support. This source is explicit and applies only to Torch; other dependencies continue to resolve from PyPI. macOS and Windows use the Torch 2.14.1 wheels from PyPI.
The dependency regression tests cover invalid PackedSequence and LSTM inputs, TorchScript annotations, and model serialization. Run them and the Sundial parameter and gradient tests from this directory in the AINode environment:
python -m unittest discover -s tests -p 'test_dependency_security.py' python -m unittest discover -s tests -p 'test_sundial_flow_loss.py'
These source tests do not validate the packaged executable or GPU support. The Cluster IT - 1C1D1A workflow builds and exercises the packaged AINode, runs the Python tests, checks the Torch wheel's sm_70 support, and executes a CUDA tensor operation on the GPU runner.