Tag
AMD and Meta contributors ported PyTorch Monarch to AMD Instinct GPUs with ROCm, enabling fault-tolerant distributed training at scale. The blog details the engineering work and validation on large clusters.