December 7, 2018
PyTorch 1.0 released
FacebookThe social network founded by Mark Zuckerberg — grew into Meta and reshaped online advertising. AI Research released PyTorch 1.0 on December 7, 2018 at NeurIPS — adding production deployment via TorchScript and a stable APIApplication programming interface — a defined way for programs to talk to each other or to a service..
What it was for
PyTorch 1.0 bridged research and production: `torch.jit.script` compiled models for C++ inferenceRunning a trained model to produce predictions — as opposed to the training phase that learns weights., and the Caffe2 backend merged into PyTorch for mobile deployment. It marked the framework's transition from academic toy to infrastructure for Meta, Tesla, and OpenAIAn AI research company — created GPT, ChatGPT, DALL-E, and the o-series reasoning models. training runs.
Why it's here
PyTorch 1.0 made the framework production-ready without sacrificing research ergonomics.
Why it mattered
TorchScript and stable APIs let teams ship models trained in notebooks to serving infrastructure.
What it solved
Researchers loved PyTorch's flexibility but could not deploy models without rewriting in C++ or TensorFlow.
Media
- ImagePyTorch
PyTorch, BSD, via Wikimedia Commons
Related
- PyTorch public beta releasedJanuary 21, 2017
- TensorFlow 2.0 releasedSeptember 30, 2019
- TensorFlow open sourcedNovember 9, 2015