models icon indicating copy to clipboard operation
models copied to clipboard
verified publisher icon intel

llama2 finetune issue on Xeon

Open huiyan2021 opened this issue 1 year ago • 2 comments

Follow the guide here: https://github.com/intel/ai-reference-models/tree/main/models_v2/pytorch/llama/training/cpu, faced several issues:

  1. https://github.com/intel/ai-reference-models/blob/main/models_v2/pytorch/llama/training/cpu/finetune.py#L36 image

  2. https://github.com/intel/ai-reference-models/blob/main/models_v2/pytorch/llama/training/cpu/finetune.py#L281 image

  3. https://github.com/intel/intel-extension-for-pytorch/issues/701

huiyan2021 avatar Sep 12 '24 05:09 huiyan2021

It's suspicious that it's due to a modeling file conflict among different transformers versions. IPEX dev is on the issue.

ZailiWang avatar Sep 13 '24 07:09 ZailiWang

Thanks for the fix, it works now except below warning which can be fixed following https://github.com/huggingface/transformers/pull/29278

image

huiyan2021 avatar Nov 20 '24 03:11 huiyan2021