fairseq2 and fairseq

Fairseq2 is the modernized successor to the original fairseq, designed to replace it with improved architecture and performance while maintaining conceptual compatibility with its predecessor's sequence-to-sequence modeling framework.

fairseq2
87
Verified
fairseq
53
Established
Maintenance 20/25
Adoption 21/25
Maturity 25/25
Community 21/25
Maintenance 2/25
Adoption 10/25
Maturity 16/25
Community 25/25
Stars: 1,122
Forks: 135
Downloads: 25,434
Commits (30d): 8
Language: Python
License: MIT
Stars: 32,190
Forks: 6,676
Downloads: —
Commits (30d): 0
Language: Python
License: MIT
No risk flags
Stale 6m No Package No Dependents

About fairseq2

facebookresearch/fairseq2

FAIR Sequence Modeling Toolkit 2

About fairseq

facebookresearch/fairseq

Facebook AI Research Sequence-to-Sequence Toolkit written in Python.

Built on PyTorch, fairseq implements diverse sequence modeling architectures—Transformers, CNNs, LSTMs, and non-autoregressive variants—with modular components for efficient distributed training via fully-sharded data parallelism. Beyond text generation, it extends to speech processing (wav2vec, speech-to-speech translation) and multimodal tasks (VideoCLIP), using Hydra for reproducible configuration management and integrating with xFormers for optimized attention mechanisms.

Related comparisons

fairseq2 and Seq2Seq-PyTorch

Scores updated daily from GitHub, PyPI, and npm data. How scores work