GPT-J benchmarking fails with IPEX 2.2.0/2.3.0

[MadhumithaSrini]

Describe the bug

• GPT-J model (source: https://huggingface.co/EleutherAI/gpt-j-6b) causes segmentation fault
• I am trying to use ipex.llm.optimize() and then apply torch.compile() on the model and then take the output using "model(input)' where input is tokenized random input

Versions

torch 2.2.0 / 2.3.0
ipex 2.2.0 / 2.3.0

[feng-intel]

Could you share your test code and platform info ?

[MadhumithaSrini]

Sure. Thanks for your response.
I am using linux platform

sample.txt

Command to run:
python sample.py

[jgong5]

@MadhumithaSrini What error did you encounter?

[MadhumithaSrini]

I encountered segmentation fault

[feng-intel]

I can reproduce the issue. Ask dev team for help. Feedback later.

[MadhumithaSrini]

How should I ask dev team for help, should I tag anyone on this issue you mean?

[feng-intel]

I will ask dev team for help internally and feedback later.

[MadhumithaSrini]

ok thank you

[feng-intel]

It will be fixed in release 2.3.100