SpeedTorch
4x faster pinned CPU -> GPU data transfer than Pytorch pinned CPU tensors, and 110x faster GPU -> CPU transfer. Augment parameter size by hosting on CPU. Use non sparse optimizers for sparse training
4x faster pinned CPU -> GPU data transfer than Pytorch pinned CPU tensors, and 110x faster GPU -> CPU transfer. Augment parameter size by hosting on CPU. Use non sparse optimizers for sparse training
Log in or sign up for Devpost to join the conversation.