对于张量列表，torchscript与torch.cat不兼容

Question

在Torchscript中使用Torch.cat时，张量列表抛出错误

这里是重现错误的最小可复制示例

import torch
import torch.nn as nn

"""
Smallest working bug for torch.cat torchscript
"""


class Model(nn.Module):
    """dummy model for showing error"""

    def __init__(self):
        super(Model, self).__init__()
        pass

    def forward(self):
        a = torch.rand([6, 1, 12])
        b = torch.rand([6, 1, 12])
        out = torch.cat([a, b], axis=2)
        return out


if __name__ == '__main__':
    model = Model()
    print(model())  # works
    torch.jit.script(model)  # throws error

预期结果将是torch.cat的torchscript输出。这是提供的错误消息：

File "/home/anil/.conda/envs/rnn/lib/python3.7/site-packages/torch/jit/__init__.py", line 1423, in _create_methods_from_stubs
    self._c._create_methods(self, defs, rcbs, defaults)
RuntimeError: 
Arguments for call are not valid.
The following operator variants are available:

  aten::cat(Tensor[] tensors, int dim=0) -> (Tensor):
  Keyword argument axis unknown.

  aten::cat.out(Tensor[] tensors, int dim=0, *, Tensor(a!) out) -> (Tensor(a!)):
  Argument out not provided.

The original call is:
at smallest_working_bug_torch_cat_torchscript.py:19:14
    def forward(self):
        a = torch.rand([6, 1, 12])
        b = torch.rand([6, 1, 12])
        out = torch.cat([a, b], axis=2)
              ~~~~~~~~~ <--- HERE
        return out

请让我知道此问题的修复程序或解决方法。

谢谢！

Answer 1

将axis更改为dim可修复错误，原始解决方案已发布here