I've discovered that OnnxBridge struggles with handling reused constants in ONNX when using SECFLOAT as the backend.
For instance, if a reshape constant (10, -1) is used in two different reshape nodes in an ONNX model, OnnxBridge will create the data 10.0, -1.0 and duplicate it in the model_weight.inp file. However, in the generated .cpp file, these values are not inputted twice, which causes an issue where all data after it cannot be read correctly.
I've discovered that OnnxBridge struggles with handling reused constants in ONNX when using SECFLOAT as the backend.
For instance, if a reshape constant (10, -1) is used in two different reshape nodes in an ONNX model, OnnxBridge will create the data 10.0, -1.0 and duplicate it in the model_weight.inp file. However, in the generated .cpp file, these values are not inputted twice, which causes an issue where all data after it cannot be read correctly.