NEML2

NEML2 is a vectorized material-modeling library that runs constitutive models on CPU and GPU using PyTorch as the tensor backend.

User guide

• Installation
• Tutorials
• Tensor types
• Physics modules
• Migration guides

Deployment & integration

• Ways to evaluate a model
• Python integration
• C++ integration
• CLI utilities
• Controlling verbosity

Contributing & development

• Building from source
• Testing
• Code quality
• Dependency pinning
• Documentation
• Jupyter notebooks
• Submitting a pull request
• Benchmarking

Reference

• py-eager — eager Python
• py-jit — in-process torch.compile
• py-aoti — compiled model from Python
• cpp-aoti — compiled model from C++
• cpp-dispatch — dispatching across devices
• cpp-eager — eager evaluation from C++
• Solvers
• Sub-batch axes
• AOTI packages
• Accelerators (device families)
• Compilation pipeline
• Syntax catalog
• Python API