News

The IBM team is combining three techniques within PyTorch – graph fusion, kernel optimizations, and parallel tensors – to achieve faster inference speeds.