Example on README throws TypeError: LlamaAttention.forward() got an unexpected keyword argument 'position_embeddings'` when executed

[mgagvani]

Running the TinyLlama Example shown in the README results in TypeError: LlamaAttention.forward() got an unexpected keyword argument 'position_embeddings' when executed.

To Reproduce

1. Run Llama script from the README

Desktop (please complete the following information):

• OS: Windows
• Torch Version: 2.4.0
• Intel NPU Acceleration Library Version: 1.3.0

[hyppo6688]

pip install transformers==4.42.4 can fix it

[hyppo6688]

The update is not timely. The transformers package is updated too quickly. It can be supported up to 4.42.4, but there will be problems with 4.43.0.

[hyppo6688]

我的python=3.9

[vunnyso]

With Run Llama script from the README.md and When using python=3.9 and using pip install transformers==4.42.4 I got below error

ModuleNotFoundError: No module named 'transformers.modeling_rope_utils'

[kv244]

Same error here. No module named modeling_rope_utils.

[Escape9002]

I tried the full example https://intel.github.io/intel-npu-acceleration-library/llm.html from the NPU Library Documentation with pip install transformers==4.42.4 and python=3.12.6 and its working.

It seems the examples from the git repo and the code from the pip install do not work together yet, at least thats what I understood from here: #108

[justindpnt]

Same issue. Any updates on this would be great! It prevents you from using the newer models.

[YadominJinta]

It has been fixed but not released, you need to install the library from source.