PyTorch 2.0

on YouTube ↗13m

" summary

PyTorch 2.0 (released March 15, 2023) delivers graph-mode speed without losing the eager UX by shipping a one‑liner torch.compile that often yields 30% gains and sometimes up to 2x. The capture layer TorchDynamo records partial graphs, inserts runtime guards and re-captures on guard failure, while the backend TorchInductor uses a PyTorch-aware IR to optimize mutations, dynamic shapes and large operator surfaces, producing the highest out‑of‑box training performance and broad model coverage.

AI-written summary · about

" chapters (4)

1. 0:00Introduction
2. 0:35Story of PyTorch 10
3. 4:12PyTorch Dynamo
4. 7:17PyTorch Inductor

" related talks

• 08mPyTorch 2.0: TorchDynamoPeng Wu · PyTorch Conference 2022
• 29mKeynoteSoumith Chintala · PyTorch Conference 2022
• 24mAdvancements in High-Performance AI/ML through PyTorch's Python CompilerSam Gross, Suraj Subramanian, Justin Jeffress · PyCon US 2023
• 15mPyTorch 2.0: Unlocking the Power of Deep Learning with the Torch Compile APIChristian Keller · Open AI + Data Forum @ Open Source Summit North America 2023
• 54mKeynote: PyTorch 2.1 Technical Deep DiveJoe Spisak, Peng Wu, Mergen Nachin, Mark Saroufim, Yanan Cao, Will Constable, Mario Lezcano · PyTorch Conference 2023
• 13mLightning Talk: Accelerated Inference in PyTorch 2.X with Torch...George Stefanakis, Dheeraj Peri · PyTorch Conference 2023
• 21mThe New Vision of Compiler-AcceleratedPeng Wu · AI @Scale 2022
• 43mPyTorch Technical Deep DivePeng Wu, Alban Desmaison, Mark Saroufim, Edward Yang · PyTorch Conference 2025