Input type (torch.cuda.DoubleTensor) and weight type (torch.cuda.FloatTensor) should be the same

HemaZ · August 8, 2018, 11:44pm

hello everyone.

i just keep getting this error whenever i try to train my model

RuntimeError: Input type (torch.cuda.DoubleTensor) and weight type (torch.cuda.FloatTensor) should be the same

you can find my code here

HemaZ/MNIST/blob/master/pytorch_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "import torch\n",
    "import torch.optim as optim\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import argparse\n",
    "from torchvision import datasets,transforms\n",
    "from torch.utils.data import Dataset, DataLoader,random_split\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt"
   ]
  },

This file has been truncated. show original

thank you

RicCu · August 9, 2018, 12:44am

Try moving the data to torch.float, i.e data.to(device=device, dtype=torch.float).

JuanFMontesinos · August 9, 2018, 12:49am

cast your input to float type by doing
input=input.float() before calling the forward pass
(I’m not sure if u should call this method after or before allocating tensor to cuda, before seems better.)

HemaZ · August 9, 2018, 12:01pm

after i tried this i got another error

RuntimeError: 1only batches of spatial targets supported (non-empty 3D tensors) but got targets of size: : [64]

JuanFMontesinos · August 9, 2018, 12:55pm

it seems you have to input 3d tensors, like x[i,j,k] but you are inputing a tensor of dim 64 ??

HemaZ · August 9, 2018, 12:56pm

yes that was a mistake I swapped (output, target) in the loss function.
thank you

jan_liao · March 11, 2019, 7:13am

i meet the same error ，if some somewhere not use cuda