Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPUgithub.com
187 points by xaskasdf 9 hours ago | 46 comments