deepmd-python-inference

DeePMD-kit model inference in Python uses the DeepPot API to load trained or pretrained models and predict energy, forces, and virial. You can execute predictions on single frames or batches directly within your Python scripts.

Yes, DeePMD-kit inference supports loading both PyTorch (.pth) and TensorFlow (.pb) trained models, as well as built-in pretrained models. The DeepPot API handles the loading process to compute energy, forces, and virial.

Model deviation analysis in DeePMD-kit computes prediction differences across multiple saved models. It is needed when comparing model performance or evaluating prediction uncertainty for atomic configurations during batch dataset testing.

Yes, DeePMD-kit Python inference supports descriptor evaluation to compute atomic environment descriptors from configurations. This runs alongside energy and force predictions using the DeepPot API on CPU or GPU backends.

Yes, installing the deepmd-kit Python package is required to run model inference, descriptor evaluation, and model deviation analysis. The package enables the DeepPot API and CLI testing workflows on CPU or GPU backends.

For batch DeePMD-kit inference on labeled datasets, use the CLI testing workflows that replicate dp test-like functionalities. This allows efficient benchmarking and batch evaluation of energy and force predictions across multiple frames.