PeopleofSuperintelligence

Builder

Ross Wightman

Also known as rwightman, wightmanr

Ross Wightman is the developer of timm, the PyTorch Image Models library that gathers image architectures, training scripts and pretrained weights for computer vision. He published timm in 2019, coauthored 'How to train your ViT?' and 'ResNet strikes back' in 2021, and joined Hugging Face in June 2022 to keep developing it. In 2022 he was one of sixteen authors of LAION-5B, whose paper received a NeurIPS 2022 Outstanding Paper award, and in 2023 he coauthored the CVPR paper on reproducible scaling laws for CLIP. He maintains timm, which reached version 1.0.30 in September 2026, and is one of four developers who lead OpenCLIP, working from Vancouver, Canada.

Links
X, LinkedIn and GitHub

Timeline

  1. 2019

    The timm library

    He publishes PyTorch Image Models, known as timm, in 2019. The open-source library collects image models, layers, optimizers, data-loaders and reference training scripts, ships with pretrained weights, and aims to reproduce ImageNet training results.

  2. 2021

    Training recipes for ViT and ResNet

    In June 2021, he coauthors 'How to train your ViT?' with Andreas Steiner, Alexander Kolesnikov, Xiaohua Zhai, Jakob Uszkoreit and Lucas Beyer, a systematic study of training data, augmentation, regularization and compute for Vision Transformers. In October 2021, he writes 'ResNet strikes back' with Hugo Touvron and Hervé Jégou; with their improved training procedure, shared in timm, a vanilla ResNet-50 reaches 80.4% top-1 accuracy on ImageNet-val without extra data or distillation.

  3. June 2022

    Joining Hugging Face

    Hugging Face announces in June 2022 that it has onboarded the developer of timm to double down on computer vision research. He continues to work on timm, both models and supporting resources, and expects tighter integration with the Hugging Face ecosystem, especially the Hub.

  4. 2022

    LAION-5B and a NeurIPS award

    He is one of sixteen authors of LAION-5B, an open dataset of 5.85 billion CLIP-filtered image-text pairs. Within the LAION project, he helps Cade Gordon debug and train the CLIP B/32 model and runs experiments on the JUWELS Booster supercomputer. The paper receives an Outstanding Paper award at NeurIPS 2022.

  5. 2023

    Reproducible scaling laws for CLIP

    At CVPR 2023, he is third author of 'Reproducible scaling laws for contrastive language-image learning', with Mehdi Cherti and Romain Beaumont as first and second authors. The study trains models on up to two billion image-text pairs from the public LAION dataset with the open-source OpenCLIP repository, identifies power law scaling for tasks such as zero-shot classification and retrieval, and releases the largest public CLIP models.

  6. April 2025

    NaFlex in timm

    In April 2025, he publishes 'NaFlex in timm' on the Hugging Face blog. The post describes his work on variable size image and aspect support in timm through the approach of SigLIP2 NaFlex, and names the data pipeline as the biggest challenge.

  7. As of October 2026

    Maintaining timm and OpenCLIP

    He maintains timm, which reaches version 1.0.30 in September 2026 with new Qwen3-VL and Sapiens2 vision transformers, and gains a hyperparameter-sweep script, sweep.py, in October 2026. He is one of four developers who lead current OpenCLIP development, with Romain Beaumont, Cade Gordon and Vaishaal Shankar, and works from Vancouver, Canada.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Hugging Face. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. The company share of AI researchers’ top papers was 31% in 2020–2024 and 8% in 1970–1974Includes Hugging Face. We estimate where the people shaping AI worked when their most-cited papers appeared: the company share was 31% in 2020–2024 and 8% in 1970–1974, and Google led all organizations in 2020–2024.
  3. Google has the highest combined h-index among AI organizationsIncludes Hugging Face. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  4. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Hugging Face. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Ross Wightman yet. Send us a free-licensed photo.

  1. GitHub - huggingface/pytorch-image-models: The largest collection of PyTorch image encoders / backbones. Including train, eval, inference, export scripts, and pretrained weights github.com. Checked 22 October 2026.
  2. [2106.10270] How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers arxiv.org. Checked 22 October 2026.
  3. [2110.00476] ResNet strikes back: An improved training procedure in timm arxiv.org. Checked 22 October 2026.
  4. Ross Whitman, developer of open source CV library Timm, joins Hugging Face analyticsindiamag.com. Checked 22 October 2026.
  5. [2210.08402] LAION-5B: An open large-scale dataset for training next generation image-text models arxiv.org. Checked 22 October 2026.
  6. NeurIPS 2022 Awards neurips.cc. Checked 22 October 2026.
  7. LAION-5B: A NEW ERA OF OPEN LARGE-SCALE MULTI-MODAL DATASETS | LAION laion.ai. Checked 22 October 2026.
  8. [2212.07143] Reproducible scaling laws for contrastive language-image learning arxiv.org. Checked 22 October 2026.
  9. NaFlex in timm huggingface.co. Checked 22 October 2026.
  10. GitHub - mlfoundations/open_clip: An open source implementation of CLIP. github.com. Checked 22 October 2026.
  11. rwightman (Ross Wightman) · GitHub github.com. Checked 22 October 2026.