92 Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU (github.com/xaskasdf) 23 hours ago xaskasdf github.com