I think you are misunderstanding just a bit. You can certainly fine-tune/train LLAMA for a specific use case. Just grab LLAMA 7B and start training, that will f

I think you are misunderstanding just a bit. You can certainly fine-tune/train LLAMA for a specific use case. Just grab LLAMA 7B and start training, that will fit on plenty of devices and you can even quantize the weights and make it run faster (and on even smaller devices). What you cannot do is use LLAMA outputs to train your ChatGPT weights.