---
id: fairscale
version: "0.4.13"
license: unclear
license_treatment: permissive
maintenance: abandoned
---

# fairscale — FairScale: A PyTorch library for large-scale and high-performance training.

License: permissive · Maintenance: abandoned · Downloads: 545.5K/mo

## What it is and what it does

FairScale is a PyTorch extension library designed to help train large models across multiple machines or GPUs by providing distributed training primitives and performance optimizations. It extends PyTorch's native capabilities with experimental features aimed at high-performance, large-scale training scenarios.

The package has been abandoned since December 2022 (1342 days without updates) and carries high install friction, likely due to compiled dependencies or tight coupling to specific PyTorch versions. While it remains permissively licensed under BSD and supports Python 3.8–3.10, the lack of maintenance means compatibility with recent PyTorch releases and Python versions is uncertain.

Use it for:

- Training large language models or vision models across multiple GPUs on a single machine.
- Distributed training workflows spanning multiple nodes in a cluster or cloud environment.
- Experimenting with advanced distributed optimization techniques in research projects using PyTorch.
- Scaling existing PyTorch models to handle datasets or model sizes that exceed single-GPU memory.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

FairScale extends PyTorch with distributed training primitives and optimizations for scaling model training across multiple machines or GPUs.

No. The package is abandoned (no updates since December 2022) with high install friction and uncertain compatibility with current PyTorch and Python versions. For distributed PyTorch training, use PyTorch's native torch.distributed or maintained alternatives like DeepSpeed or Hugging Face Accelerate instead.

## Install

pip install fairscale
uv add fairscale
poetry add fairscale

## Installing fairscale

Before you install: High install friction; the package is abandoned (last release 2022-12-11, 1342 days ago) and has no runtime dependencies listed, suggesting it may require manual compilation or system-level PyTorch setup. No recent maintenance signals.

License in practice: BSD License (permissive) places no restrictions on commercial or private use, though you should verify the exact license terms given the empty license_raw field.

Quickstart:

pip install fairscale==0.4.13
import fairscale
# Use fairscale distributed training utilities with PyTorch models

Requires PyTorch installation and compatible CUDA/system libraries; package is abandoned and may not work with recent PyTorch versions.

Verify before relying:

- Compatibility with PyTorch versions released after the package's final release in December 2022.
- Whether the package builds successfully on modern Python 3.10+ environments.
- Specific distributed training features and APIs available in version 0.4.13.

## Package facts

- License: not declared (permissive)
- Python support: supports_current
- Install friction: high
- Maintenance: abandoned
- Downloads: 545.5K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

pytorch distributed training, large scale model training, multi-gpu training library, pytorch scaling utilities, distributed deep learning, pytorch performance optimization, multi-node training framework, distributed-training, pytorch-extension, abandoned

[View on SkillFed](https://skillfed.io/packages/fairscale) · [View on PyPI](https://pypi.org/project/fairscale/)