llama-7b is nowhere even close to GPT-4. Not even llama-65b is.

We'll see if fine tuning can improve this, but e.g. alpaca-30b is still inferior to ChatGPT-3.5.