Provides research-grade implementations and pretrained models for sequence tasks (translation, LM, speech). Offers multi-GPU training, fast generation (beam/sampling/lexical constraints), mixed-precision, and state sharding — aimed at researchers reproducing or extending papers.
Provides differentiable, GPU-accelerated computer-vision operators and geometric building blocks on top of PyTorch; includes 500+ ops, augmentation pipelines, and pre-trained models for detection, matching, and segmentation—suitable for research and production vision pipelines.
