Closed Andrea8Testa closed 3 years ago
@ louie-tsai Can you look at this and work with the customer on it.
@jingxu10 - Can you take a look at this please
Which version of AI Analytics Toolkit have you tried? How did you installed it? I just tried ai kit docker image (https://hub.docker.com/r/intel/oneapi-aikit), the hello world script worked correctly.
I tried the Intel® oneAPI AI Analytics Toolkit (https://software.intel.com/content/www/us/en/develop/tools/oneapi/ai-analytics-toolkit.html#gs.2chtw8) for Linux operating system selecting the Web & Local distribution. I chose the online installer and I used the two commands shown in the Command Line Installation section. After the installation I followed this guide (https://software.intel.com/content/www/us/en/develop/documentation/get-started-with-ai-linux/top/before-you-begin.html#before-you-begin) for the set up. All the scripts are executed without errors, so I thought that the installation was executed correctly. The hello world script seems not to be included in this toolkit, so I had to clone it from this Github Repository. I tried to clone it inside and outside the environment "pytorch" but the result remained the same.
I cannot reproduce the error you reported. The code ran successfully with the following steps on my machine.
Ok thank you, I tried your steps and it actually works. Maybe I did something wrong when I followed the get started guide (https://software.intel.com/content/www/us/en/develop/documentation/get-started-with-ai-linux/top.html).
Anyway I am still not able to run this code on the GPU, that is the real reason why I downloaded this toolkit (since I have an Intel graphic card).
If I add to the code the lines:
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
It still not able to find any GPU.
Intel GPU support has not been enabled in public version of oneAPI package yet. The packages in current oneAPI AI Analytics Toolkit are all for CPU for now. We will release the packages later once they are ready to go.
Ok, so it seems there are no options to use GPU with Intel graphic card. Anyway, thank you very much for the help.
Following the tutorial to make this sample work, after the installation of the oneAPI AI Analytics Toolkit, I executed the commands:
I get the error:
Segmentation fault (core dumped)
Instead, if I try to clone the environment:
conda create --name usr_pytorch --clone pytorch
the error is:
How can be solved?