Input list is zero for ONNX model

Question

I have exported my model from pytorch to ONNX but I am getting empty input list on running this code

input_names = [input.name for input in onnx_model.graph.input]

My model is

    TaggingAgent(
  (_encoder): BiGraphEncoder(
    (_utt_encoder): BiRNNEncoder(
      (_word_embedding): Embedding(2770, 128)
      (_rnn_cell): LSTM(128, 128, batch_first=True, bidirectional=True)
    )
    (_dialog_layer_user): GAT(
      (firstlayer): GraphAttentionLayer (256 -> 256)
      (secondlayer): GraphAttentionLayer (256 -> 256)
      (attention_0): GraphAttentionLayer (256 -> 256)
      (attention_1): GraphAttentionLayer (256 -> 256)
      (attention_2): GraphAttentionLayer (256 -> 256)
      (attention_3): GraphAttentionLayer (256 -> 256)
      (attention_4): GraphAttentionLayer (256 -> 256)
      (attention_5): GraphAttentionLayer (256 -> 256)
      (attention_6): GraphAttentionLayer (256 -> 256)
      (attention_7): GraphAttentionLayer (256 -> 256)
      (out_att): GraphAttentionLayer (2048 -> 256)
    )
  )
  (_decoder): RelationDecoder(
    (_sent_layer_dict): ModuleDict(
      (0): BiLSTMLayer(
        (_rnn_layer): LSTM(256, 128, batch_first=True, bidirectional=True)
      )
      (1): UniLinearLayer(
        (_linear_layer): Linear(in_features=256, out_features=256, bias=True)
      )
    )
    (_act_layer_dict): ModuleDict(
      (0): BiLSTMLayer(
        (_rnn_layer): LSTM(256, 128, batch_first=True, bidirectional=True)
      )
      (1): UniLSTMLayer(
        (_rnn_layer): LSTM(256, 256, batch_first=True)
      )
    )
    (_relate_layer): GraphRelation(
      (_sent_linear): Linear(in_features=256, out_features=256, bias=False)
      (_act_linear): Linear(in_features=256, out_features=256, bias=False)
      (_dialog_layer): GAT(
        (firstlayer): GraphAttentionLayer (256 -> 256)
        (secondlayer): GraphAttentionLayer (256 -> 256)
        (attention_0): GraphAttentionLayer (256 -> 256)
        (attention_1): GraphAttentionLayer (256 -> 256)
        (attention_2): GraphAttentionLayer (256 -> 256)
        (attention_3): GraphAttentionLayer (256 -> 256)
        (attention_4): GraphAttentionLayer (256 -> 256)
        (attention_5): GraphAttentionLayer (256 -> 256)
        (attention_6): GraphAttentionLayer (256 -> 256)
        (attention_7): GraphAttentionLayer (256 -> 256)
        (out_att): GraphAttentionLayer (2048 -> 256)
      )
    )
    (_sent_linear): Linear(in_features=256, out_features=3, bias=True)
    (_act_linear): Linear(in_features=256, out_features=15, bias=True)
  )
  (_criterion): NLLLoss()
)

What am I doing wrong here? I need some elements in the input list to work further. Is the problem with my pytorch model or with conversion from pytorch to onnx. What to make change?

Input list is zero for ONNX model

0 Answers0