Closed chadcwilliams closed 3 months ago
Here's the reason why: configuration = { 'device': torch.device("cuda" if torch.cuda.is_available() else "cpu") if not ddp else torch.device("cpu") }
That's because for DDP Training the CPU is hosting the whole process and distributing to each available GPU
Recommendation: Would close the issue
Here's the reason why: configuration = { 'device': torch.device("cuda" if torch.cuda.is_available() else "cpu") if not ddp else torch.device("cpu") }
That's because for DDP Training the CPU is hosting the whole process and distributing to each available GPU
Recommendation: Would close the issue