We fine-tuned Llama 405B on AMD GPUs | Hacker News