How to use the content of the notice

[Evfidiw]

Thanks for your interesting work.
Could you provide more detailed guidance on the content of Notice? Should I change LlamaForCausalLM to AutoModel on line163 in finetune_kg.py? But this will give the error.
In addition, could you please provide the memory of your A800? I cannot run on a GPU with 48GB of memory.

[luoxindi]

You can modify the code to:
'''
model = AutoModelForCausalLM.from_pretrained(
base_model,
#load_in_8bit=True,
torch_dtype=torch.float32,
device_map="auto",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
'''
And comment out the code: shutil.copyfile(base_model...)
The memory of A800 is about 80G. And I recommend that you adjust the mini_batch_size to 2 or 1. But this may affects the final results.