Keynote
summary
PyTorch 1.3 announces roughly 1,200 contributors (about 50% YOY growth) and a push from research into production at Facebook where most models now train on PyTorch. Key concrete features include three quantization modes—post-training quantization, quantization-aware training and dynamic quantization—yielding 2–4× inference speedups, named tensors for clearer dimension handling, new libraries like Krypton for privacy, Captum for explainability, Detectron2 and FairSeq, TPU GA and broad cloud support including Azure, AWS and Alibaba, plus a Portal pose detection use case.
AI-written summary · about
chapters (5)
related talks
54mKeynote: PyTorch 2.1 Technical Deep Dive
12mKeynote: How PyTorch Became the Foundation of the AI Revolution
06mKeynote: PyTorch Lightning: Powering the GenAI Revolution from Research to the...
43mPyTorch Technical Deep Dive
12mLightning Talk: State of PyTorch
06mCommunity Updates
29mKeynote
19mIncreasing the Impact of AI Through Better Software (Pytorch Updates)
