Installation

  • ÆÄÀ̽㠹öÁ¯ üũ (Python version Check)
  • ÆÄÀÌÅäÄ¡ ¼³Ä¡ (PyTorch Installation)
  • Äí´Ù ¹× CuDNN üũ (Cuda & CuDNN Check)

1. Python Version Check

ÆÄÀ̽㠹öÁ¯ üũ

In [1]:
import sys
print(sys.version)

3.6.8 (default, Jan 14 2019, 11:02:34) 
[GCC 8.0.1 20180414 (experimental) [trunk revision 259383]]

2. PyTorch Installation

  • ±¸±Û Äݶóºê ¹öÁ¯¿¡ µû¶ó ÆÄÀÌÅäÄ¡°¡ ¼³Ä¡µÇ¾î ÀÖÀ»¼öµµ ÀÖ°í ¾Æ´Ò ¼öµµ ÀÖ½À´Ï´Ù.
  • ¼³Ä¡°¡ ¾ÈµÇ¾î ÀÖÀ» °æ¿ì ¾Æ·¡¿Í °°Àº ¸í·É¾î·Î ¼³Ä¡ÇÏ¸é µË´Ï´Ù.
  • !pip3 install torch torchvision
In [0]:
import torch
In [3]:
!pip3 install torch torchvision

Requirement already satisfied: torch in /usr/local/lib/python3.6/dist-packages (1.1.0)
Requirement already satisfied: torchvision in /usr/local/lib/python3.6/dist-packages (0.3.0)
Requirement already satisfied: numpy in /usr/local/lib/python3.6/dist-packages (from torch) (1.16.4)
Requirement already satisfied: pillow>=4.1.1 in /usr/local/lib/python3.6/dist-packages (from torchvision) (4.3.0)
Requirement already satisfied: six in /usr/local/lib/python3.6/dist-packages (from torchvision) (1.12.0)
Requirement already satisfied: olefile in /usr/local/lib/python3.6/dist-packages (from pillow>=4.1.1->torchvision) (0.46)

3. Cuda & cudnn Version Check

  • ÆÄÀÌÅäÄ¡¸¦ ÅëÇØ °¢°¢ ¸î ¹öÁ¯ÀÌ ¼³Ä¡ µÇ¾îÀÖ´ÂÁö È®ÀÎÇØÁÝ´Ï´Ù.
In [4]:
import torch

print("Torch version:{}".format(torch.__version__))
print("cuda version: {}".format(torch.version.cuda))
print("cudnn version:{}".format(torch.backends.cudnn.version()))

Torch version:1.1.0
cuda version: 10.0.130
cudnn version:7501

4. Command Line cuda & cudnn Check

  • Äí´Ù ¹× CuDNN ¹öÁ¯Àº Ä¿¸Çµå¶óÀÎ ¸í·É¾î·Îµµ È®ÀÎ °¡´ÉÇÕ´Ï´Ù.
In [5]:
!nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
In [6]:
!cat /usr/include/x86_64-linux-gnu/cudnn_v*.h | grep CUDNN_MAJOR -A 2

#define CUDNN_MAJOR 7
#define CUDNN_MINOR 6
#define CUDNN_PATCHLEVEL 2
--
#define CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)

#include "driver_types.h"

5. PyTorch CPU & GPU Tensor Check

  • ÆÄÀÌÅäÄ¡ ÅÙ¼­¸¦ »ý¼ºÇغ½À¸·Î½á Á¦´ë·Î ¼³Ä¡ µÇ¾ú´ÂÁö, Àß µ¿ÀÛÇÏ´ÂÁö È®ÀÎÇØÁÝ´Ï´Ù.

5-1 Create CPU tensor

In [7]:
# https://pytorch.org/docs/stable/torch.html?highlight=tensor#torch.tensor
# 0À¸·Î Â÷ÀÖ´Â 2x3 ÇüÅÂÀÇ ÅÙ¼­¸¦ »ý¼ºÇÕ´Ï´Ù.
cpu_tensor = torch.zeros(2,3)
print(cpu_tensor)

tensor([[0., 0., 0.],
        [0., 0., 0.]])

5-2 Allocate tensor on GPU

In [8]:
# https://pytorch.org/docs/stable/tensor_attributes.html#torch.torch.device
# ¾î´À ÀåÄ¡(cpu ȤÀº gpu)¿¡ ÅÙ¼­¸¦ ¿Ã¸±Áö ÁöÁ¤ÇÕ´Ï´Ù.
# ¾Æ·¡´Â torch.device¶ó´Â ÇÔ¼ö¸¦ »ç¿ëÇØ gpu·Î ÀåÄ¡¸¦ ÁöÁ¤ÇÕ´Ï´Ù. 
device = torch.device('cuda')

# https://pytorch.org/docs/stable/cuda.html?highlight=available#torch.cuda.is_available
# gpu°¡ »ç¿ë °¡´ÉÇÑÁö È®ÀÎÇØÁÝ´Ï´Ù.
if torch.cuda.is_available():
  
  # https://pytorch.org/docs/stable/tensors.html?highlight=#torch.Tensor.to
  # cpu¿¡ ÀÖ¾ú´ø ÅÙ¼­¸¦ to ÇÔ¼ö¸¦ ÀÌ¿ëÇØ ÁöÁ¤ÇسõÀº ÀåÄ¡(¿©±â¼­´Â gpu)·Î ¿Ã·ÁÁÝ´Ï´Ù.
  gpu_tensor = cpu_tensor.to(device)
  print(gpu_tensor)

tensor([[0., 0., 0.],
        [0., 0., 0.]], device='cuda:0')

5-3 Reallocate tensor back on CPU

In [9]:
# device ÇÔ¼ö¿Í to ÇÔ¼ö¸¦ ÀÌ¿ëÇØ gpu¿¡ ÀÖ´ø ÅÙ¼­¸¦ ´Ù½Ã cpu·Î ¿Å°Ü¿Ã ¼ö ÀÖ½À´Ï´Ù.
cpu_tensor_back = gpu_tensor.to(torch.device('cpu'))
cpu_tensor_back
Out[9]:
tensor([[0., 0., 0.],
        [0., 0., 0.]])
In [0]: