2Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU (opens in new tab)(github.com)395xaskasdf1mo ago101
