This is a cleaned-up version of decapoda-research/llama-7b-hf. It should be ready to be used right away. Somehow you need LlamaTokenizer to be able to load the tokenizer, it still doesn't work with AutoTokenizer

from transformers import LlamaTokenizer, AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("kz919/llama_7b")
tokenizer = LlamaTokenizer.from_pretrained("kz919/llama_7b")