tensorplay.func.functional_call

tensorplay.func.functional_call(module: Module, parameter_and_buffer_dicts: dict[str, Any] | Sequence[dict[str, Any]], args: Any | tuple | None = None, kwargs: dict[str, Any] | None = None, *, tie_weights: bool = True, strict: bool = False)

Runs module with the parameters and buffers given, not its own.

The module’s own state is put back afterwards, including if module raises, so this is safe to call on a live model.

Parameters:

• module (tensorplay.nn.Module) – the module to call.

• parameter_and_buffer_dicts (dict or sequence of dicts) – the state to substitute, keyed by the names named_parameters and named_buffers report. Several dicts are merged; overlapping keys are an error, since which one wins would be arbitrary.

• args (Any or tuple) – positional arguments for the module. A non-tuple value is passed as the single argument.

• kwargs (dict) – keyword arguments for the module.

• tie_weights (bool) – when the module ties two names to one tensor, keep them tied by requiring the replacement to be shared as well. Default: True.

• strict (bool) – reject names that the module does not have. Default: False.

Example

>>> params = dict(model.named_parameters())
>>> functional_call(model, params, (x,))