Skip to yearly menu bar Skip to main content


Moshpit SGD: Communication-Efficient Decentralized Training on Heterogeneous Unreliable Devices

Max Ryabinin · Eduard Gorbunov · Vsevolod Plokhotnyuk · Gennady Pekhimenko

Keywords: [ Federated Learning ] [ Deep Learning ] [ Optimization ]


Training deep neural networks on large datasets can often be accelerated by using multiple compute nodes. This approach, known as distributed training, can utilize hundreds of computers via specialized message-passing protocols such as Ring All-Reduce.However, running these protocols at scale requires reliable high-speed networking that is only available in dedicated clusters.In contrast, many real-world applications, such as federated learning and cloud-based distributed training, operate on unreliable devices with unstable network bandwidth.As a result, these applications are restricted to using parameter servers or gossip-based averaging protocols.In this work, we lift that restriction by proposing Moshpit All-Reduce — an iterative averaging protocol that exponentially converges to the global average.We demonstrate the efficiency of our protocol for distributed optimization with strong theoretical guarantees.The experiments show 1.3x speedup for ResNet-50 training on ImageNet compared to competitive gossip-based strategies and 1.5x speedup when training ALBERT-large on preemptible compute nodes.

Chat is not available.