Hi, when initializing a Tensor using a numpy compatible scalar with an explicit dtype, the resulting numpy() call returns the wrong data type:

import numpy as np
from onnx_ir import DataType, Tensor

tensor = Tensor(value=np.float32(0.5), dtype=DataType.FLOAT)
print(tensor.numpy().dtype) # outputs float64 instead of float32

Potential fix

Maybe pass the dtype of _raw here