Is 32GB RAM the sweet spot for running Llama 7B and 13B models?

Yes, 32GB is the sweet spot. For 7B in 4-bit, 16GB works if you close other apps, but you’ll hit swapping. For 13B in 4-bit, 16GB is borderline—32GB gives you headroom for weights, context, and other apps, meaning faster inference and longer prompts. Future You will thank you.