‘tuple‘ object has no attribute ‘log_softmax‘

问题描述

pytorch 运行深度学习代码，F.cross_entropy 出现问题：

Traceback (most recent call last):
  File "main.py", line 217, in <module>
    acc = test_clean()
  File "main.py", line 201, in test_clean
    loss = criterion(outputs, targets)
  File "/home/kaiyuan/anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
    result = self.forward(*input, **kwargs)
  File "/home/kaiyuan/anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/modules/loss.py", line 961, in forward
    return F.cross_entropy(input, target, weight=self.weight,
  File "/home/kaiyuan/anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/functional.py", line 2468, in cross_entropy
    return nll_loss(log_softmax(input, 1), target, weight, None, ignore_index, None, reduction)
  File "/home/kaiyuan/anaconda3/envs/torch/lib/python3.8/site-packages/torch/nn/functional.py", line 1605, in log_softmax
    ret = input.log_softmax(dim)
AttributeError: 'tuple' object has no attribute 'log_softmax'

原因分析

一般来说，这个 criterion 前面跟的都是前向传播的过程，这个前向传播的输出一般都是 outputs。但是在有些情况下，输出不仅是outputs。比如 inception-v3 输出还有个 aux，即输出为 (outputs, aux)，当之前的输出仍然用outputs来接的话，相当于收到的是一个 tuple，再送进 criterion 就出错了。所以这个问题一般是前面 feed forward 的问题，和 criterion 无关。

解决办法

在之前多放一个变量接就好

---

参考链接

1. stackoverflow某个帖子
2. github某个issue