cuDNN 错误：CUDNN_STATUS_BAD_PARAM。有人可以解释为什么我收到此错误以及如何纠正它？

Question

我正在尝试使用 Pytorch 实现字符 LSTM。但是我收到了 cudnn_status_bad_params 错误。这是训练循环。我在 line output = model(input_seq) 上遇到错误。

for epoch in tqdm(range(epochs)):
  for i in range(len(seq)//batch_size):
   sidx = i*batch_size
   eidx = sidx + batch_size
   x = seq[sidx:eidx]
   x = torch.tensor(x).cuda()
   input_seq =torch.nn.utils.rnn.pack_padded_sequence(x,seq_lengths,batch_first = True)
   y = out_seq[sidx:eidx]
   output = model(input_seq)
   loss = criterion(output,y)
   loss.backward()
   optimizer.step()
/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py in __call__(self, *input, **kwargs)
    487             result = self._slow_forward(*input, **kwargs)
    488         else:
--> 489             result = self.forward(*input, **kwargs)
    490         for hook in self._forward_hooks.values():
    491             hook_result = hook(self, input, result)   
/usr/local/lib/python3.6/dist-packages/torch/nn/modules/rnn.py in forward(self, input, hx)
    180         else:
    181             result = _impl(input, batch_sizes, hx, self._flat_weights, self.bias,
--> 182                            self.num_layers, self.dropout, self.training, self.bidirectional)
    183         output = result[0]
    184         hidden = result[1:] if self.mode == 'LSTM' else result[1]

 RuntimeError: cuDNN error: CUDNN_STATUS_BAD_PARAM

Answer 1

我遇到了同样的错误，如果你切换到 CPU，你会得到更好的错误描述。就我而言，问题出在我提供给网络的输入类型上。我正在发送我猜long，而模型需要float. 我进行了以下更改并且代码有效。基本上切换到 CPU 会给出更好的错误描述。

input_seq = input_seq.float().cuda()