Training Foundation Models on a Full-Stack AMD Platform | Dark Hacker News