Python

Source: user_tensors/PythonTensor.py

Tensor built from an inline Python expression.

The expr option is evaluated in a namespace pre-populated with torch, every public name from neml2.types (Scalar, SR2, SSR4, free functions), math, and np (numpy, if importable). Cross-references to other [Tensors] entries resolve by bare identifier: writing base is equivalent to tensor('base'), which avoids HIT’s restriction on nested quotes.

The expression’s value (torch.Tensor or a TensorWrapper subclass) is returned verbatim; the call site (typically neml2.model.Model.declare_typed_parameter mode 2) is responsible for wrapping a raw tensor into a typed wrapper if needed.

Other options

expr — str · required

Python expression producing the tensor value. A multi-line block must assign its result to a variable named result.