# pytorch/gloo

Collective communications library with various primitives for multi-machine training.

Repository: https://github.com/pytorch/gloo
Canonical: https://ross.abutalabs.com/products/pytorch-gloo
Homepage: https://pytorch.org/docs/stable/distributed.html
Language: C++
License: NOASSERTION
License Family: other
Topics: collectives, distributed-training, pytorch
Last push: 2026-08-21T18:55:40+00:00

## Health v2 (maintenance only)
Score: 76/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 35, longevity 100
- inputs: {"age_days": 3499, "days_push": 12, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1447, forks 363 (observed 2026-08-28T04:04:45.361201+00:00)

## What it is
Gloo is a C++ collective communications library providing primitives like barrier, broadcast, and allreduce for multi-machine machine learning training. It abstracts transport over IP or InfiniBand (with GPUDirect support) and integrates with PyTorch's distributed package.

## Use cases
- run distributed training across multiple machines
- perform allreduce synchronization between training workers
- broadcast tensors to cluster nodes
- accelerate GPU-to-GPU transfers with InfiniBand and GPUDirect
- use torch.distributed backend on CPU-only clusters
- coordinate rendezvous between machines via Redis or MPI

## When to choose
- you need a CPU-friendly distributed backend for PyTorch
- your cluster uses IP or InfiniBand transports
- you need collective algorithms like allreduce or broadcast in C++

## When to avoid
- you need cutting-edge GPU collectives (prefer NCCL)
- you need active development of new features (Gloo is maintenance-only)
- you are on Windows or macOS (Gloo targets Linux)

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, gpu-computing, networking
- domain: machine-learning, microservices, deep-learning
- platform: cpp
- tags: collective-communications, distributed-training, allreduce, infiniband, nccl, pytorch, linux, gpu

## Member repositories
- pytorch/gloo (main) score 76

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:45.361201+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:36:07.940281+00:00, confidence not recorded.
  - readme: https://github.com/pytorch/gloo (fetched 2026-08-28T04:04:45.361201+00:00, sha b0191b2a846a)
  - homepage: https://pytorch.org/docs/stable/distributed.html (fetched 2026-08-29T11:45:51.642161+00:00, sha 01dea697ee0b)
  - site_page: https://docs.pytorch.org/docs/2.13/distributed.html (fetched 2026-08-29T11:45:51.644624+00:00, sha 4fe6c00b4529)
- Data as of 2026-08-30T08:39:29.467469+00:00.