Building machine learning systems in 2019 feels like stacking together Lego. Except, you have to construct all the Lego pieces from scratch. In the dark.

You open up a database of pictures used to train artificial intelligence systems. At first, things seem straightforward. You’re met with thousands of images: apples and oranges, birds, dogs, horses, mountains, clouds, houses, and street signs. But as you probe further into the dataset, people begin to appear: cheerleaders, scuba divers, welders, Boy Scouts, fire walkers, and flower girls. Things get strange: A photograph of a woman smiling in a bikini is labeled a “slattern, slut, slovenly woman, trollop.”

Anyone in the industry knows how confusing it can be to discern between a data scientist, data engineer, model ops engineer, research scientist, and the list goes on. Titles aside, getting models into production (i.e. serving a ML model that provides predictions) is frankly where the money is.

There are many barriers to success in data-driven initiatives. Chief among them is the difficulty many organizations face in operationalizing analytics: deploying, monitoring, and managing analytics and AI in business processes.

In 2008, the World Wildlife Fund ran a campaign that used pixelation to represent the number of animals left for endangered species. One pixel represents an animal, so an image appears more pixelated when there are fewer animals left.

There are not enough hours in the day for one person to do everything alone.

In my experience, every nontrivial machine learning project is eventually stitched together with bug-ridden and unmaintainable internal tools.

Transformers provides state-of-the-art general-purpose architectures (BERT, GPT-2, RoBERTa, XLM, DistilBert, XLNet…) for Natural Language Understanding (NLU) and Natural Language Generation (NLG). The library glues these together through an abstraction layer that speeds up experimentation.