TracerWarning: Converting a tensor to a Python index 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!

I convert a pytorch model to onnx.

example = torch.rand(10, 3, 224, 224)

torch.onnx.export(net,               # model being run
                  example,                         # model input (or a tuple for multiple inputs)
                  "./infer/tsm_resnet50.onnx",   # where to save the model (can be a file or file-like object)
                  export_params=True,        # store the trained parameter weights inside the model file
                  opset_version=10,          # the ONNX version to export the model to
                  do_constant_folding=True,  # whether to execute constant folding for optimization
                  input_names = ['input'],   # the model's input names
                  output_names = ['output'], # the model's output names
                #   operator_export_type=torch.onnx.OperatorExportTypes.ONNX_ATEN_FALLBACK,
                  dynamic_axes={'input' : {0 : 'batch_size'},    # variable lenght axes
                                'output' : {0 : 'batch_size'}})

And then it show me that: this is my log file log.txt the problem snippet:

            out = torch.zeros_like(x)
            out[:, :-1, :fold] = x[:, 1:, :fold]  # shift left
            out[:, 1:, fold: 2 * fold] = x[:, :-1, fold: 2 * fold]  # shift right
            out[:, :, 2 * fold:] = x[:, :, 2 * fold:]  # not shift

how can replace it ? thanks my version: python3.6 torch 1.2.0 torchvison 0.4.0

[yesid-acm]

Anyone explain it, please? it´s an issue for prediction ? how to solve it?

[miketrimmel]

I´m facing the same warning with Python=3.7.9, pytorch=1.6.0, onnxruntime=1.4.0, onxruntime-tools=1.4.2 when converting a bert model.

def export_onnx_bert_model(model, onnx_model_path, max_seq_len):
    with torch.no_grad():
        inputs = {"input_ids":      torch.ones(1, max_seq_len, dtype=torch.int64),
                  "attention_mask": torch.ones(1, max_seq_len, dtype=torch.int64),
                  "token_type_ids": torch.ones(1, max_seq_len, dtype=torch.int64)}
        
        outputs = model(**inputs)
        symbolic_names = {0: "batch_size", 1: "max_seq_len"}
        torch.onnx.export(model,                                            
                          (inputs["input_ids"],                             
                           inputs["attention_mask"],
                           inputs["token_type_ids"]),                      
                          onnx_model_path,                                  
                          opset_version=11,                                 
                          do_constant_folding=True,                        
                          input_names=['input_ids',                         
                                       'input_mask',
                                       'segment_ids'],
                          output_names=['output'],                         
                          dynamic_axes={'input_ids': symbolic_names,
                                        'input_mask' : symbolic_names,
                                        'segment_ids' : symbolic_names})
        print("ONNX Model exported to {0}".format(onnx_model_path))

export_onnx_bert_model(bert_model, "bert.onnx", MAX_SEQ_LEN)

TracerWarning: Converting a tensor to a Python index 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! position_ids = self.position_ids[:, :seq_length]

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! input_tensor.shape == tensor_shape for input_tensor in input_tensors