All Tools
F
Fine-tuningFreeOpen Source
FAIRSEQ2
FAIR sequence modeling toolkit for training modern models
MIT
ABOUT
The original fairseq stack aged out of current PyTorch and hardware, so FAIR research recipes were hard to reproduce. Fairseq2 rebuilds sequence modeling with a cleaner recipe API, published model cards, and installable CUDA/CPU wheels so teams can train and evaluate speech and LLM workloads.
INSTALL
pip install fairseq2INTEGRATION GUIDE
1. Train or evaluate a FAIR speech or translation recipe with pip-installed fairseq2
2. Load published Fairseq2 model cards for inference without rebuilding fairseq
3. Prototype a custom sequence model on top of Fairseq2's PyTorch recipe system
4. Reproduce Meta FAIR training setups on CUDA or CPU wheels
TAGS
pythonpytorchfairseqspeechtranslationtrainingmeta