This is a great question. My best answer is that there's a speed/intelligence trade-off. The smaller weights (7B) will run faster and require less memory, but y

This is a great question. My best answer is that there's a speed/intelligence trade-off. The smaller weights (7B) will run faster and require less memory, but you won't get the same quality responses as the 13B / 70B model. I think there may be a Llama 30B variant coming soon too.