GTC Silicon Valley-2019: Training AI Models Faster With Distributed Training in PyTorch

GTC Silicon Valley-2019 ID:S9830:Training AI Models Faster With Distributed Training in PyTorch

Soumith Chintala(Facebook),Teng Li(Facebook)
In addition to the new production deployment-oriented capabilities included in the 1.0 release of PyTorch, the deep learning framework also added improved distributed training, allowing researchers and developers to easily parallelize computations across processes and clusters of machines. The PyTorch dev team at Facebook has continued to up the performance, and will be walking through new benchmarks, and how developers can readily take advantage of distributed training in PyTorch and NVIDIA GPUs to train their models faster.

Slides part 1(pdf),Slides part 2(pdf)