how to register custom operator with attributes of a list?

[CapJunkrat]

In the provided documents, it showed an example of custom operator with attribute "padding_length", which has type int64. (code listed below.) What if this attribute is a list? How should I modify the example code to accommadate this change? Since attribute with list type is pretty common, such as the kernel of a conv op, I assume the list type attribute is already supported.

@onnx_op(op_type="GPT2Tokenizer",
            inputs=[PyCustomOpDef.dt_string],
            outputs=[PyCustomOpDef.dt_int64, PyCustomOpDef.dt_int64],
            attrs={"padding_length": PyCustomOpDef.dt_int64})
def bpe_tokenizer(s, **kwargs):
    padding_length = kwargs["padding_length"]
    input_ids, attention_mask = cls.tokenizer.tokenizer_sentence([s[0]], padding_length)
    return input_ids, attention_mask

[CapJunkrat]

@wenbingl I found that in the function cast_attributes, only int, float and string types are supported. Is it true that list of ints/floats is not supported?

[wenbingl]

@wenbingl I found that in the function cast_attributes, only int, float and string types are supported. Is it true that list of ints/floats is not supported?

Yes, if they are vector, they can be input as an ONNX initializer, i.e. a tensor-like type.

[CapJunkrat]

Yes, if they are vector, they can be input as an ONNX initializer, i.e. a tensor-like type.

@wenbingl What if the operator attribute need to be a vector? Since it is quite common to have vector attributes(such as conv kernel & paddings), will this feature be supported soon?