[torchlib] torch.ops.aten.embedding_bag.padding_idx may see padding_idx=None

[justinchuby]

ERR_export: Failed to convert the exported program to an ONNX model. [96mThis is step 3/3[0m of exporting the model to ONNX. Next steps:- If there is a missing ONNX function implement it and register it to the registry.- If there is an internal error during ONNX conversion debug the error and summit a PR to PyTorch.- Create an error report with torch.onnx.export(... report=True) and save the ExportedProgram as a pt2 file. Create an issue in the PyTorch GitHub repository against the [96m*onnx*[0m component. Attach the error report and the pt2 model.## Exception summary<class 'AssertionError'>: padding_idx must not be None. This is likely a dispatcher error<class 'torch.onnx._internal.exporter._errors.GraphConstructionError'>: Error when calling function 'TracedOnnxFunction(<function aten_embedding_bag_padding_idx at 0x7f890955be20>)' with args '[SymbolicTensor('p_model_emb_l_0_weight' type=Tensor(FLOAT16) shape=[100000064] producer=None index=None) SymbolicTensor('args_2_0' type=Tensor(INT64) shape=[100] producer=None index=None) SymbolicTensor('select' type=Tensor(INT64) shape=[1] producer=node_Gather_5 index=0) False 0 True None False None]' and kwargs '{}'<class 'torch.onnx._internal.exporter._errors.ConversionError'>: Error when translating node %embedding_bag : [num_users=1] = call_function[target=torch.ops.aten.embedding_bag.padding_idx](args = (%p_model_emb_l_0_weight %args_2_0 %select False 0 True None False None) kwargs = {}). See the stack trace for more information.(Refer to the full stack trace above for more information.)

[justinchuby]

cc @titaiwangms @shubhambhokare1 @xadupre

[titaiwangms]

I recall we had this conversation before: https://github.com/pytorch/pytorch/blob/35ff5084e6a5bbf7c897840943ee3ac846ffaaf1/aten/src/ATen/native/native_functions.yaml#L2161

Vaguely, it seems that rounding_mode is an optional attribute. It could be provided with str or None, or not even provided.

[titaiwangms]

That's why in symbolic_opset9: it's *args

https://github.com/pytorch/pytorch/blob/35ff5084e6a5bbf7c897840943ee3ac846ffaaf1/torch/onnx/symbolic_opset9.py#L410

[justinchuby]

Here it's optional: https://github.com/pytorch/pytorch/blob/35ff5084e6a5bbf7c897840943ee3ac846ffaaf1/aten/src/ATen/native/native_functions.yaml#L2346C1-L2346C237

[titaiwangms]

Oops, I was talking about ate.div