fairseq
Python toolkit from Meta FAIR for training sequence models like translation and speech systems; archived.
- Stars
- 32.2k
- Forks
- 6.7k
- Licence
- MIT
- Language
- Python
- Last update
- 1.1 years ago
- Type
- Framework
This repository is archived and no longer maintained.
What it is
Fairseq is a Python toolkit from Facebook AI Research for building sequence models that turn one sequence of text or audio into another, such as translation, summarization, and language modeling. It includes reference implementations of many published research papers, from early convolutional and LSTM models to Transformers and speech models like wav2vec. The repository is now archived, so it no longer receives updates.
- Train a custom machine translation model on your own parallel text
- Reproduce a paper's results using the example folders linked from the README
- Experiment with self-supervised speech models such as wav2vec 2.0
- Fine-tune or evaluate released models such as the MMS speech models mentioned in the README
Great for
- Researchers reproducing published sequence-modeling papers
- ML engineers who want reference implementations of Transformer-style models
- Teams training custom translation or speech models in PyTorch
- Students studying how models like RoBERTa or wav2vec are built
Not ideal for
- Anyone wanting a ready-made app, since this is a training toolkit with no end-user interface
- Teams that need active maintenance, because the repository is archived and last pushed in September 2025
- Beginners without Python and machine learning background, since the README assumes researcher-level knowledge
How to get it
- 1
Clone and build from source
Clone the repository
git clone https://github.com/facebookresearch/fairseq - 2
Run it
- Clone the repository using the command above
- Read the main README for the list of implemented papers and their linked example folders
- Open the example README for the paper or task closest to your goal
- Follow that example's instructions, since the install and run steps for building from source are not shown in the README excerpt reviewed here
- 3
Get the most out of it
- Use the Hydra configuration framework the README says was adopted in November 2020 for configuring new and existing projects
- Look at the fully sharded data parallel docs to shard parameters and optimizer state across GPUs
- Check the xFormers integration noted in the README for efficient attention options
- Start from released models such as the MMS speech models or RoBERTa rather than training from scratch
Downloads
Strengths
- Covers a wide range of published sequence-modeling papers in one codebase
- Released under the permissive MIT license
- Includes per-paper example documentation that links to the original research
Trade-offs
- The repository is archived, so it will not get new fixes or features
- The README is long and points to many separate example guides, which can be hard to navigate for newcomers
- Setup and usage are aimed at people already familiar with PyTorch and research workflows
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| fairseq (this) | 32.2k | Framework | advanced | 1.1 years ago | Researchers reproducing published sequence-modeling papers |
| OpenHands | 92.5k | App | intermediate | 4 days ago | Developers who want coding agents running on a server while their laptop is off |
| spaCy | 34k | Library | intermediate | 10 days ago | Python developers building text-processing features into an application |
| semantic-kernel | 28.7k | Library | intermediate | 4 days ago | Developers already using Azure OpenAI or OpenAI who want a structured SDK for agents |
| SuperAGI | 17.7k | AI model / agent | advanced | 1.8 years ago | Python developers experimenting with autonomous AI agents |
| nano-vllm | 16.3k | Library | intermediate | 6 months ago | Developers who want to understand how LLM inference engines work internally |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.