$ pip3 install onnx $ PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python python3 import onnx key = 'value' attr_type = onnx.onnx_cpp2py_export.defs.OpSchema.AttrType.TENSOR data = ...

We've been using python-dwca-reader with no problems loading about 13k occurrences. We now need to scale it up to load about 3.25m occurrences. Changing the code from: core_df = ...