# pip install git+https://github.com/huggingface/transformers.git@main accelerate
from transformers import LlamaTokenizer, AutoModelForCausalLM

tokenizer = LlamaTokenizer.from_pretrained("./")
model = AutoModelForCausalLM.from_pretrained("./")

inputs = tokenizer("A cat sat", return_tensors="pt")["input_ids"]
outputs = model.generate(inputs, max_new_tokens=5)
print(tokenizer.decode(outputs[0]))