关于DDP 的问题

Hi, blogger! Your open source project has been a great help to me, but I recently ran into a problem with DDP. I'm using an A6000 graphics card, and there's no problem when running with a single card. But when I use multiple cards, both deepspeed and torchrun commands always get stuck. I checked it myself, and it seems that there's a bottleneck in the data distribution stage, but I don't know how to fix it. I was wondering if anyone has encountered this problem and how they solved it.