Following the open-sourcing of image processing library Spectrum in January, natural language processing modeling framework PyText late last year, and AI reinforcement learning platform Horizon in November, Facebook’s AI research division announced that Pythia, a modular plug-and-play framework that enables data scientists to quickly build, reproduce, and benchmark AI models, is now freely available on GitHub.
As Facebook explains in a blog post, Pythia — which is built atop the company’s PyTorch machine learning framework — is principally intended for vision and language tasks, such as answering questions related to visual data and automatically generating image captions.
It incorporates elements of Facebook AI Research’s top entries in AI competitions like LoRRA, a vision and language model that won both the VQA Challenge 2018 and Vizwiz Challenge 2018, and it’s capable of showing how previous state-of-the-art AI systems achieved top benchmark results and comparing their performance to that of new models.
Modules with implementations for commonly-used vision and language layers are present and accounted for, as is support for distributed training; built-in corpora including VQA, VizWiz, TextVQA, and VisualDialog; and features like multitasking, which allows training on multiple corpora together. We hope that removing some of the obstacles will allow researchers to more quickly develop new ways for people and intelligent machines to communicate.
Read more here
