horovod/horovod

Health v2 · maintenance only

Flags: archived no_license

Adoption not part of the score

14688 stars · 2233 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Horovod is a distributed deep learning training framework for TensorFlow, Keras, PyTorch, and Apache MXNet, originally developed at Uber. It lets users scale single-GPU training scripts to run across many GPUs with minimal code changes, using MPI-style allreduce and integration with Spark and Ray. The project is now inactive.

Use cases

• scale a single-GPU PyTorch training script to multiple GPUs
• run distributed TensorFlow training across a cluster
• train deep learning models on Spark clusters
• use NCCL-based allreduce for fast multi-node GPU training
• unify data processing and model training in one pipeline

When to choose

• you need multi-GPU or multi-node data-parallel training with PyTorch, TensorFlow, Keras, or MXNet
• you want minimal code changes to scale an existing training script
• you are maintaining an existing Horovod-based training pipeline

When to avoid

• starting a new project, since Horovod is officially inactive and unmaintained
• you only need single-GPU training
• you prefer framework-native distributed training like torch.distributed or tf.distribute

Facets

library · maturity abandoned

machine-learning deep-learning llm-training machine-learning deep-learning microservices gpu-computing python cloud distributed-training mpi nccl tensorflow pytorch keras mxnet spark data-parallelism linux gpu docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="horovod/horovod")