onnx_light.onnx.tools#

Public helpers that complement onnx_light.onnx.

onnx_light.onnx.tools.make_random_input(elem_type: int, shape: list[int], seed: int) → ndarray#

Creates a random NumPy array for an ONNX tensor input.

Uses the same deterministic pseudo-random generators as the onnx-light runtime (onnx_light.onnx_lib.backend.random).

Parameters:

  • elem_type – TensorProto data-type integer (e.g. TensorProto.FLOAT).

  • shape – Concrete non-negative integer dimensions.

  • seed – Integer seed forwarded to the random generator.

Returns:

A numpy.ndarray of the appropriate dtype and shape.

Raises:

NotImplementedError – For unsupported element types (e.g. STRING).