RuntimeError:Input and parameter tensors are not at the same device, found input tensor at cuda:0 an

错误原因: 输入x和输出y(或模型参数)存放的位置不同所产生的

这种错误主要是因为输入x和输出y(或模型参数)存放的位置不同所产生的
如果你是错误1: 输入x在cuda(gpu)中， 模型参数在cpu中
想把输入x放入到gpu中，一般就是找到输入参数x，然后再调用使用参数x之前添加一行代码x.to(device)(其中device=“cuda”)
如果你是错误2: 输入x在cpu中， 模型参数cuda(gpu)在中
找到定义model的代码,在定义的后面添加一行代码 model.to(device)
具体操作如下:

错误1:RuntimeError: Input and parameter tensors are not at the same device, found input tensor at cuda:0 and parameter tensor at cpu

1.1 输入x在cuda(gpu)中， 模型参数在cpu中

测试代码demo: :
此时输入x在gpu, 但是model存放在cpu中 所以运行如下代码会报错误1

import torch
import torch.nn as nn
from torch.nn import LSTM
device = "cuda" if torch.cuda.is_available() else "cpu"     # 有gpu用gpu, 没有就用cpu
x = torch.Tensor([[1,2,3], [2,3,4]])  # x shape (2,3)  (seq_len, 词向量维度)
class Testmodel(nn.Module):
    def __init__(self, input_dim, lstm_layer, lstm_hidden_dim, dropout):
        super(Testmodel, self).__init__()
        self.lstm_encoding = LSTM(input_dim, num_layers=lstm_layer, hidden_size=lstm_hidden_dim,
                                  dropout=0.5)  #
    def forward(self, x: torch.Tensor):
        output, (hn, cn) = self.lstm_encoding(x)
        return output

model = Testmodel(
    input_dim=3,
    lstm_layer=2,
    lstm_hidden_dim=4,
    dropout=0.5,
)

# 此时输入x在gpu, 但是model存放在cpu中 所以会报错
x = x.to(device)    # 将x放入到gpu内存中
output = model(x)  # 调用forward方法 x (2,3) lstm 输入维度3, 输出维度4,
print(output)   # output shape (2,4)

1.2 解决方法

方法1:直接注释下面代码,将输入x放入到cpu内存中和输出保持一致

x = x.to(device)    # 将x放入到gpu内存中

方法2(推荐):添加一行代码model.to(device),将模型的参数放入到gpu中，和输入x位置保持一致，改完后案例代码如下

import torch
import torch.nn as nn
from torch.nn import LSTM
device = "cuda" if torch.cuda.is_available() else "cpu"     # 有gpu用gpu, 没有就用cpu
x = torch.Tensor([[1,2,3], [2,3,4]])  # x shape (2,3)  (seq_len, 词向量维度)
class Testmodel(nn.Module):
    def __init__(self, input_dim, lstm_layer, lstm_hidden_dim, dropout):
        super(Testmodel, self).__init__()
        self.lstm_encoding = LSTM(input_dim, num_layers=lstm_layer, hidden_size=lstm_hidden_dim,
                                  dropout=0.5)  #
    def forward(self, x: torch.Tensor):
        output, (hn, cn) = self.lstm_encoding(x)
        return output

model = Testmodel(
    input_dim=3,
    lstm_layer=2,
    lstm_hidden_dim=4,
    dropout=0.5,
)
model.to(device)  # !!!!!!!!!!!!新添加的代码在这里
# 此时输入x在gpu, 但是model在gpu中 错误解决!!!!!
x = x.to(device)    # 将x放入到gpu内存中
output = model(x)  # 调用forward方法 x (2,3) lstm 输入维度3, 输出维度4,
print(output)   # output shape (2,4)

错误2.RuntimeError: Input and parameter tensors are not at the same device, found input tensor at cpu and parameter tensor at cuda:0

2.1 输入在cpu中， 输出(模型参数)在cuda(gpu)中

测试代码demo:
此时输入x在cpu, 但是model存放在gpu中 所以运行如下代码会报错误2

import torch
import torch.nn as nn
from torch.nn import LSTM
device = "cuda" if torch.cuda.is_available() else "cpu"     # 有gpu用gpu, 没有就用cpu
x = torch.Tensor([[1,2,3], [2,3,4]])  # x shape (2,3)  (seq_len, 词向量维度)
class Testmodel(nn.Module):
    def __init__(self, input_dim, lstm_layer, lstm_hidden_dim, dropout):
        super(Testmodel, self).__init__()
        self.lstm_encoding = LSTM(input_dim, num_layers=lstm_layer, hidden_size=lstm_hidden_dim,
                                  dropout=0.5)  #
    def forward(self, x: torch.Tensor):
        output, (hn, cn) = self.lstm_encoding(x)
        return output

model = Testmodel(
    input_dim=3,
    lstm_layer=2,
    lstm_hidden_dim=4,
    dropout=0.5,
)
model.to(device)    # 将模型参数放到gpu中
# 此时输入x在cpu, 但是model参数在gpu中 所以会报错

output = model(x)  # 调用forward方法 x (2,3) lstm 输入维度3, 输出维度4,
print(output)   # output shape (2,4)

2.2 解决方法

方法1:找到代码model.to(device)直接注释, 将模型参数放入到cpu内存中和输入x位置保持一致

model.to(device)    # 将模型参数放到gpu中

方法2(推荐):添加一行代码x = x.to(device) ,将模型的参数放入到gpu中，和输入x位置保持一致，改完后案例代码如下

import torch
import torch.nn as nn
from torch.nn import LSTM
device = "cuda" if torch.cuda.is_available() else "cpu"     # 有gpu用gpu, 没有就用cpu
x = torch.Tensor([[1,2,3], [2,3,4]])  # x shape (2,3)  (seq_len, 词向量维度)
class Testmodel(nn.Module):
    def __init__(self, input_dim, lstm_layer, lstm_hidden_dim, dropout):
        super(Testmodel, self).__init__()
        self.lstm_encoding = LSTM(input_dim, num_layers=lstm_layer, hidden_size=lstm_hidden_dim,
                                  dropout=0.5)  #
    def forward(self, x: torch.Tensor):
        output, (hn, cn) = self.lstm_encoding(x)
        return output

model = Testmodel(
    input_dim=3,
    lstm_layer=2,
    lstm_hidden_dim=4,
    dropout=0.5,
)
model.to(device)  # 将模型参数放入到gpu内存中
# 此时输入x在gpu, 同时model参数也在gpu中 不会报错
x = x.to(device)    # !!!!!!!!!!!!新添加的代码在这里
output = model(x)  # 调用forward方法 x (2,3) lstm 输入维度3, 输出维度4,
print(output)   # output shape (2,4)

本人水平有限， 如有错误欢迎指正交流