AttributeError: 'LlamaConfig' object has no attribute 'rope_theta'

[lvtao65535]

I followed the instructions in the README to deploy tinychat on jetson agx orin and tried to get the llama-2-7b-chat model. However, Meta doesn't seem to provide a model in huggingface format, so I downloaded llama-2-7b-chat-hf uploaded by another user as an alternative.
When running the command in step 4 “Run the TinyChat demo”, I encountered the following error:

Traceback (most recent call last):
  File "/SSD/codes/./llm-awq/tinychat/demo.py", line 175, in <module>
    model = model_type_dict["llama"](config).half()
  File "/SSD/codes/llm-awq/tinychat/models/llama.py", line 325, in __init__
    self.model = Transformer(params)
  File "/SSD/codes/llm-awq/tinychat/models/llama.py", line 281, in __init__
    self.layers.append(TransformerBlock(layer_id, params))
  File "/SSD/codes/llm-awq/tinychat/models/llama.py", line 250, in __init__
    self.self_attn = LlamaAttentionFused(args)
  File "/SSD/codes/llm-awq/tinychat/models/llama.py", line 85, in __init__
    self.rope_theta = args.rope_theta
  File "/SSD/anaconda3/envs/tinychat-test/lib/python3.10/site-packages/transformers/configuration_utils.py", line 261, in __getattribute__
    return super().__getattribute__(key)
AttributeError: 'LlamaConfig' object has no attribute 'rope_theta'

I searched for this error and found a similar case, where it seems that I need to upgrade the transformers library to a higher version.
Even though this contradicts the earlier instructions specifying transformers==4.32.0, I tried pip install --upgrade transformers to upgrade transformers to version 4.45.1, and was able to run the demo after that.
Although the problem has been solved for now, I still want to ask if this is normal and whether it will cause protential problems? Is there a better solution?

[Louym]

Thank you for your report! We have also noticed this error and upgraded the version of transformers in #218.