AI People Wiki

Builders & open source

Natalia Gimelshein

Core Maintainer, PyTorch

  • Core Maintainer, PyTorch (open-source project)1

Natalia Gimelshein is a PyTorch core maintainer and a maintainer of PyTorch's CUDA support, according to PyTorch's maintainers page, which also lists Gimelshein as an emeritus maintainer of NestedTensor and of the profiling and monitoring tools. Gimelshein co-authored the 2019 PyTorch paper, listed there with NVIDIA, and the PyTorch 2 paper (ASPLOS 2024), listed with OpenAI. At PyTorch Conference Europe 2026, Gimelshein took part in the "Meet the Developers" session for PyTorch module maintainers.1234

At a glance

Current role
Core Maintainer
Company
PyTorch
Work listed
2
Sources cited
9
Profile reviewed

Research

Two of Natalia Gimelshein's earlier papers addressed GPU efficiency. Gimelshein co-authored vDNN (2016), a runtime memory manager that lets deep neural networks train beyond a GPU's memory capacity by virtualizing memory across the GPU and CPU, and with Maxim Milakov wrote Online normalizer calculation for softmax (2018), which computes the standard softmax with fewer memory accesses.56

Gimelshein is a co-author of the 2019 PyTorch paper, listed there with NVIDIA, and of the ASPLOS 2024 PyTorch 2 paper on TorchDynamo and TorchInductor, listed with OpenAI. Gimelshein also appears among the contributors to the October 2024 GPT-4o System Card.237

Open-source work

PyTorch's governance page lists Gimelshein as one of the project's core maintainers and as a maintainer of its CUDA support, and as an emeritus maintainer of NestedTensor and of the profiling and monitoring tools. At PyTorch Conference Europe 2026, Gimelshein joined the "Meet the Developers" session for module maintainers.14

Work and contributions

  • PyTorch (CUDA) Open-source code

    Listed as a PyTorch core maintainer and a CUDA maintainer; co-author of "PyTorch: An Imperative Style, High-Performance Deep Learning Library" (2019). The repository is under a BSD-style license.189

  • PyTorch 2: TorchDynamo and TorchInductor (torch.compile) Published research

    Co-author of the ASPLOS 2024 paper introducing TorchDynamo and TorchInductor, which implement torch.compile in PyTorch 2.3

Timeline

  1. Co-authored vDNN, on virtualized memory for training large neural networks on GPUs.5
  2. Co-authored Online normalizer calculation for softmax with Maxim Milakov.6
  3. Co-authored "PyTorch, An Imperative Style, High-Performance Deep Learning Library".8
  4. Co-authored the PyTorch 2 paper on TorchDynamo and TorchInductor at ASPLOS 2024.3
  5. Listed among the contributors to the GPT-4o System Card.7
  6. Took part in the "Meet the Developers" session at PyTorch Conference Europe 2026.4

Other documented roles

Roles the sources document that are not confirmed as current.

  • Listed with OpenAI as an author of the PyTorch 2 paper (2024), OpenAI3
  • Listed with NVIDIA as an author of the 2019 PyTorch paper, NVIDIA2

Where they work

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

9 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. PyTorch Governance | Maintainers (PyTorch 2.14 documentation)PyTorch · Documentation · published Jun 5, 2026 · accessed Oct 21, 2026
  2. PyTorch: An Imperative Style, High-Performance Deep Learning Library, paper PDF with author affiliations (arXiv 1912.01703)arXiv · Paper · published Dec 3, 2019 · accessed Oct 21, 2026
  3. PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation (ASPLOS '24)PyTorch · Paper · published Apr 2024 · accessed Oct 21, 2026
  4. PyTorch Conference Europe 2026: A Landmark Moment for Open Source AI in ParisPyTorch · Announcement · published Apr 15, 2026 · accessed Oct 21, 2026
  5. vDNN: Virtualized Deep Neural Networks for Scalable, Memory-Efficient Neural Network Design (arXiv 1602.08124)arXiv · Paper · published Feb 25, 2016 · accessed Oct 21, 2026
  6. Online normalizer calculation for softmax (arXiv 1805.02867)arXiv · Paper · published May 8, 2018 · accessed Oct 21, 2026
  7. GPT-4o System Card (arXiv 2410.21276)arXiv · Paper · published Oct 25, 2024 · accessed Oct 21, 2026
  8. PyTorch: An Imperative Style, High-Performance Deep Learning Library (arXiv 1912.01703)arXiv · Paper · published Dec 3, 2019 · accessed Oct 21, 2026
  9. pytorch/pytorch LICENSE (BSD-style license)PyTorch (GitHub repository) · License · accessed Oct 21, 2026