brendondgr
commented
5 months ago Addenum - Windows 10 So I wanted to add that whenever I use the Windows 10 instructions and create a conda environment with all of the requirements (proper torch, torchvision and torchaudio versions), it seems to work fine. I had also installed the versions of this that relate to the most recent release of Intel extension for pytorch. However, whenever I add a new library into the mix (such as torchmetrics, monai, etc), it seems to either downgrade, upgrade, or completely remove the previously mentioned requirements.
Here is the error whenever I add in the new libraries, specifically seems to highlight the importing of torch. As I stated, everything was working fine until I had decided to install matplotlib:
import warnings
warnings.filterwarnings("ignore")
# Import Torch, MONAI and other libraries
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision.transforms import Compose, Normalize, ToTensor
from torch.utils.data import DataLoader
# from torch.utils.tensorboard import SummaryWriter
from monai.transforms import Compose, ScaleIntensity, EnsureChannelFirst, Resize, EnsureChannelFirstd
from monai.networks.layers import Norm
from monai.networks.nets import UNet
from monai.losses import DiceLoss
from monai.inferers import sliding_window_inference
import os
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
# Set random seed
torch.manual_seed(0)
# Import custom classes/functions
from cg_Dataset import load_dataSpecifically the error is the following:
OSError: [WinError 127] The specified procedure could not be found. Error loading "..\anaconda3\envs\dl_bu\lib\site-packages\torch\lib\backend_with_compiler.dll" or one of its dependencies.
vishnumadhu365
Describe the issue
The Issue I am having: When attempting to import this library, I am getting the following error:
import intel_extension_for_pytorch as ipexInstructions I followed for download: https://intel.github.io/intel-extension-for-pytorch/index.html#installation?platform=gpu&version=v2.1.30%2bxpu&os=linux%2fwsl2&package=pip
Here are the version in case you don't want to read through it:
From these, this may be overkill, but I added the following paths to .bashrc:
Other Solutions I have Attempted: I have also activated OneAPI environment as well as the conda environment prior to attempting the import, but this results in a different error. Here is the order of commands (Note, conda env "dl" contains all mentioned libaries):
source /home/bdgr/intel/oneapi/setvars.shconda activate dlpython -c "import torch; import intel_extension_for_pytorch as ipex;"This import (after activating OneAPI) results in the following Error:
I don't know what this error is. The other one I knew had something to do with OneAPI, but any help would be appreciated! I have also tried to find if there was a GPU/XPU in a Jupyter notebook, but it was not detected.
Specifications: Operating System: Fedora Workstation 40 Graphics Card: Intel Arc A770 CPU: Ryzen 5 3600 Motherboard: MSI B450-A Pro MAX (Updated to Latest Kernel version to support Re-Bar)