Llama 3.1 8B specs, VRAM requirements, and which GPUs can run it. The go-to small model for local inference.