Progressive Layer Dropping

DeepSpeed Sparse Attention

Training a Trillion Parameters with Pipeline Parallelism

Up to 5x less communication and 3.4x faster training through 1-bit Adam

DeepSpeed with 1-bit Adam: 5x less communication and 3.4x faster training

10x bigger model training on a single GPU with ZeRO-Offload

Powering 10x longer sequences and 6x faster execution through DeepSpeed Sparse Attention

DeepSpeed Microsoft Research Webinar is now on-demand Permalink

DeepSpeed Microsoft Research Webinar on August 6th, 2020 Permalink

Microsoft DeepSpeed achieves the fastest BERT training time

ZeRO-2 & DeepSpeed: Shattering Barriers of Deep Learning Speed & Scale Permalink

An Order-of-Magnitude Larger and Faster Training with ZeRO-2

The Fastest and Most Efficient BERT Training through Optimized Transformer Kernels

Turing-NLG: A 17-billion-parameter language model by Microsoft Permalink

DeepSpeed was used to train the world’s largest language model.

ZeRO & DeepSpeed: New system optimizations enable training models with over 100 billion parameters Permalink

Developed by Microsoft AI & Research.