SlowLlama: Finetune llama2-70B and codellama on MacBook Air without quantization
https://github.com/okuvshynov/slowllama

https://news.ycombinator.com/item?id=37796863