Hi! 
Update: I tried using the custom DeBERTaV2 ONNXConfig copied from this link on the HF github. This piece of code essentially looks like this:
class DebertaV2OnnxConfig(OnnxConfig):
@property
def inputs(self) -> Mapping[str, Mapping[int, str]]:
if self.task == "multiple-choice":
dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
else:
dynamic_axis = {0: "batch", 1: "sequence"}
if self._config.type_vocab_size > 0:
return OrderedDict(
[("input_ids", dynamic_axis), ("attention_mask", dynamic_axis), ("token_type_ids", dynamic_axis)]
)
else:
return OrderedDict([("input_ids", dynamic_axis), ("attention_mask", dynamic_axis)])
@property
def default_onnx_opset(self) -> int:
return 12
def generate_dummy_inputs(
self,
preprocessor: Union["PreTrainedTokenizerBase", "FeatureExtractionMixin"],
batch_size: int = -1,
seq_length: int = -1,
num_choices: int = -1,
is_pair: bool = False,
framework: Optional["TensorType"] = None,
num_channels: int = 3,
image_width: int = 40,
image_height: int = 40,
tokenizer: "PreTrainedTokenizerBase" = None,
) -> Mapping[str, Any]:
dummy_inputs = super().generate_dummy_inputs(preprocessor=preprocessor, framework=framework)
if self._config.type_vocab_size == 0 and "token_type_ids" in dummy_inputs:
del dummy_inputs["token_type_ids"]
return dummy_inputs
After running this along with the mentioned steps in the reference (in the OG post), on the export step, I encounter the following error message:
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:561: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
q_ids = np.arange(0, query_size)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:561: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
q_ids = np.arange(0, query_size)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:562: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
k_ids = np.arange(0, key_size)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:562: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
k_ids = np.arange(0, key_size)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:566: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
rel_pos_ids = torch.tensor(rel_pos_ids, dtype=torch.long)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:695: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
scale = math.sqrt(query_layer.size(-1) * scale_factor)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:749: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
).repeat(query_layer.size(0) // self.num_attention_heads, 1, 1)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:751: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
query_layer.size(0) // self.num_attention_heads, 1, 1
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:770: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
scale = math.sqrt(pos_key_layer.size(-1) * scale_factor)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:782: TracerWarning: Converting a tensor to a Python float might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
scale = math.sqrt(pos_query_layer.size(-1) * scale_factor)
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:783: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
if key_layer.size(-2) != query_layer.size(-2):
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py:112: TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
output = input.masked_fill(rmask, torch.tensor(torch.finfo(input.dtype).min))
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-40-694ff91713ca> in <module>()
4 onnx_config,
5 onnx_config.default_onnx_opset,
----> 6 onnx_path)
9 frames
/usr/local/lib/python3.7/dist-packages/transformers/models/deberta_v2/modeling_deberta_v2.py in symbolic(g, self, mask, dim)
133 to_i=sym_help.cast_pytorch_to_onnx["Byte"],
134 )
--> 135 output = masked_fill(g, self, r_mask, g.op("Constant", value_t=torch.tensor(torch.finfo(self.dtype).min)))
136 output = softmax(g, output, dim)
137 return masked_fill(g, output, r_mask, g.op("Constant", value_t=torch.tensor(0, dtype=torch.uint8)))
AttributeError: 'torch._C.Value' object has no attribute 'dtype'
OOF, this error makes no sense to me, can someone help me out with this? What’s going on? 
)