Error when loading pretrained OpenLlama model

[kevin-guimard-ext]

System Info

• transformers version: 4.38.2
• Platform: Windows-10-10.0.19044-SP0
• Python version: 3.10.13
• Huggingface_hub version: 0.21.3
• Safetensors version: 0.4.2
• Accelerate version: 0.27.2
• Accelerate config: not found
• PyTorch version (GPU?): 2.2.1+cu118 (True)
• Tensorflow version (GPU?): 2.10.1 (True)
• Flax version (CPU?/GPU?/TPU?): not installed (NA)
• Jax version: not installed
• JaxLib version: not installed
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: no

Who can help?

@ArthurZucker @younesbel

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

When running this snippet:

from transformers import OpenLlamaForCausalLM
model = OpenLlamaForCausalLM.from_pretrained("openlm-research/open_llama_7b")

I get the following error:

You are using a model of type llama to instantiate a model of type open-llama. This is not supported for all configurations of models and can yield errors.
...
AttributeError: 'OpenLlamaAttention' object has no attribute 'rope_theta'

Expected behavior

The code snippet has been taken from the documentation (https://huggingface.co/docs/transformers/model_doc/open-llama), it is expected to work.

[ArthurZucker]

I can reproduce. This model is deprecated, feel free to open a PR for a fix, otherwise would recommend using Llama, we no longer maintain OpenLlama

[kevin-guimard-ext]

Ok, I'll switch to Llama then. I wanted to try OpenLlama because I heard that Llama required a subscription, but I may be wrong.

[ArthurZucker]

No, OpenLlama is basically Llama but with some "optimisations". You should be able to load https://huggingface.co/NousResearch/Llama-2-7b-hf 🤫

[kevin-guimard-ext]

It's a pity that the "Run the model" section has not been written for this model, with a snippet explaining how to use it. Must the weights be downloaded from the Meta repo?