onnx_light.onnx_core#
Python facade for the onnx_core C++ library.
This module mirrors the C++ onnx_core namespace at the Python level,
providing the shared types and graph-manipulation helpers that sit below both
onnx_op and onnx_shapes in the dependency hierarchy.
The TensorType enum and ToTypeString helper are the canonical Python
spelling for the element / sequence type descriptor defined in
onnx_proto/type_helper.h. onnx_light.onnx_op.TensorType re-exports
the same object for backward compatibility.
The three graph-analysis functions (collect_external_inputs,
collect_node_inputs, collect_remaining_inputs) mirror the helpers
defined in onnx_core/graph/graph_manipulations.h and are also re-exported
from onnx_light.onnx for convenience.
- class onnx_light.onnx_core.TensorType(*values)#
Element or sequence tensor type supported by onnx-light.