3 ms·40 GB wouldn't even be enough for LLaMA 65B with 8 bit quantization. Let alone something like GPT-4.by cubefox 3y ago40 GB wouldn't even be enough for LLaMA 65B with 8 bit quantization. Let alone something like GPT-4.