How can I run llama2 from this repo?

[dnatarajan00]

I want to be able to run both llama2 (7b) and llama3 (8b) from this repo to be able to easily compare them. I know I can use the old repo to run llama2 but the sku_list.py file makes it seem like it should be possible to run llama2 here too.

I have tried manually changing the tokenizer files and did all of the following:

• set CHECKPOINT_DIR to a llama2 checkpoint
• set the tokenizer.py and tokenizer.model files to the llama2 versions (and added a get_instance method to the llama2 tokenizer file pointing to the correct tokenizer.model file)
• (HACK) set model_args.vocab_size = tokenizer.n_words

But it then failes with the error:

File "llama_models/llama3/api/chat_format.py", line 54, in init
self.vision_token = self.tokenizer.special_tokens["<|image|>"]
AttributeError: 'Tokenizer' object has no attribute 'special_tokens'

What is the correct way to run llama2 here?