Intel® Developer Cloud for the Edge - Bare-metal - Missing OneAPI Libraries

[JasonPua]

I have a customer interested in testing out IPEX with Pytorch on Intel Developer Cloud for the Edge on the bare metal instance. 

It seems there is a node option with the config:

Intel Xeon D Processor

Intel Xeon D processor D-2796NT (formerly code named Ice Lake)

65 GB

Intel® Data Center GPU Flex 170 (Discrete)

None

CPU: d-2796nt GPU: intel-data-center-gpu-flex-170

idc084

The IPEX library was installed with "pip"
# python3 -m pip install torch==1.13.0a0+git6c9b55e torchvision==0.14.1a0 intel_extension_for_pytorch==1.13.120+xpu -f https://developer.intel.com/ipex-whl-stable-xpu

Installation was successful with "pip show torch"

=========================================================================

Name: torch

Version: 1.13.0a0+git6c9b55e

Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration

Home-page: https://pytorch.org/

Author: PyTorch Team

Author-email: packages@pytorch.org

License: BSD-3

Location: /home/u183618/.local/lib/python3.8/site-packages

Requires: typing-extensions

Required-by: accelerate, optimum, torchvision, triton

=========================================================================

However once we tried importing torch on terminal with [python3 -c "import torch"], we are getting error that mkl libraries are missing.
=========================================================================
OSError: libmkl_intel_lp64.so.2: cannot open shared object file: No such file or directory

=========================================================================

Upon further checking I can see that the node idc084 has no OneAPI root as well.

ls /opt/intel/oneapi/  

also returns missing error. 

Does anyone from developer cloud team for the edge know if we can install OneAPI on those nodes? 

[JesusE_Intel · Accepted solution]

Hi JasonPua,

The Intel Developer Cloud for the Edge nodes do not have the Intel oneAPI Software Installed. If IPEX only requires pip packages you could try to setup a customer kernel with the required packages and submit the job to the node. If you need additional software outside pip packages I recommend using DevCloud for oneAPI.

Please take a look at this discussion where I provided instructions to setup a custom kernel:

https://community.intel.com/t5/Intel-DevCloud/Unable-to-import-pandas-after-activating-openvino-2022-2-0/m-p/1479334#M8003

Regards,

Jesus

[JasonPua]

Hi there Jesus, 

I've read through the documentation for IPEX and MKL libraries are certainly one of the prerequisites to run some AI Model training on the FLEX170 GPU. I've tried installing mkl via "pip install mkl" and exported the library_path to the MKL libraries but no cigar. So its certainly an outside pip packages situation. 

Thank you for the pointers on the custom kernels though, that sure will come in useful. 

Best Regards, 
Jason

[JesusE_Intel]

Hi JasonPua,

Any time! Please let me know if I can proceed to close this discussion. or if you have any additional questions.

Regards,

Jesus

[JasonPua]

Yes sir, none for now.

Please proceed to close the discussion. 

Thank you, 
Jason

[JesusE_Intel]

If you need any additional information, please submit a new question as this thread will no longer be monitored.