TorchText | PyTorch Developer Day 2020

In this video, research scientist George Zhang explains efforts to a) accelerate NLP research by providing reusable, orthogonal, and correct building blocks for cutting-edge research based on the knowledge of the text domain and research communities; and b) provide a solution to transfer from research to production: we integrate those pipeline and modules with a wide range of PyTorch capabilities, such as torchscript, quantization, distributed data parallel, and mobile. With those goals in mind, we provide easy access to some commonly used datasets, text processing pipelines and transforms, like tokenizer and vocab, and some NLP related modules. The text domain team in PyTorch wants to develop a good technical understanding in the NLP area and build new research collaborations.