facebookresearch/fairseq2
A PyTorch-based sequence modeling toolkit for training custom language and generation models.
Not currently ranked — collecting fresh signals.
star history
fairseq2 is a research-focused sequence modeling library that provides modular APIs for training custom models on content generation tasks including language modeling, speech translation, and multilingual speech recognition. It is a reboot of the original fairseq with a clean architecture and supports recent FAIR research on Large Concept Models, Seamless speech translation, and diverse preference optimization.
Frequently asked
- What is facebookresearch/fairseq2?
- A PyTorch-based sequence modeling toolkit for training custom language and generation models.
- Is fairseq2 open source?
- Yes — facebookresearch/fairseq2 is open source, released under the MIT license.
- What language is fairseq2 written in?
- facebookresearch/fairseq2 is primarily written in Python.
- How popular is fairseq2?
- facebookresearch/fairseq2 has 1.1k stars on GitHub.
- Where can I find fairseq2?
- facebookresearch/fairseq2 is on GitHub at https://github.com/facebookresearch/fairseq2.