Function call_op

pub fn call_op(
    op_name: impl AsRef<str>,
    overload: impl AsRef<str>,
    args: &[RValue],
    kwargs: &HashMap<String, RValue>,
    flatten_results: bool,
) -> Result<Vec<RValue>, CallOpError>

Call a PyTorch-dispatched operator by name.

op_name is the fully qualified name of the operator, like "aten::add".

overload is the name of the overload, like "Scalar". Due to a quirk of libtorch, the default overload must be called by passing an empty string.

flatten_results is a flag that indicates whether the results of the operator should be flattened into a single list. Extracting out values from lists, tuples and dicts recursively.

Errors

If the called operator throws an exception, a cxx::Exception will be returned which contains the C++ exception.