Thomas Wolf

Career

Thomas Wolf's first career was in physics: Wolf worked on laser-plasma interactions at the BELLA Center of Lawrence Berkeley National Laboratory and earned a PhD in statistical and quantum physics at Sorbonne University and ESPCI. Wolf then worked as a patent attorney at the IP firm Cabinet Plasseraud for five years. Consulting for deep learning startups around 2015 led Wolf to machine learning, and about a year later Wolf joined the founding of Hugging Face.¹

At Hugging Face, where Wolf is co-founder and Chief Science Officer, Wolf's site describes involvement from the start in the company's open-source, open-science and robotics efforts. Wolf's current research interest is whether AI systems can help produce new scientific knowledge.¹²

Open-source work

Wolf's site lists the Transformers, Datasets, Diffusers, Accelerate, DataTrove, smolagents and LeRobot libraries among those Wolf created or advised. Wolf made the first commits to the Transformers repository in October 2018 and the Accelerate repository in October 2020.¹³⁶

Wolf has also worked on open-science projects such as the BigScience Workshop, which led to the BLOOM model and dataset, the FineWeb dataset and the Ultra-Scale Playbook. Wolf is among the authors of the July 2025 SmolLM3 announcement.¹⁷

Publications

Wolf is first author of the 2019 Transformers paper, which Wolf's site lists as an EMNLP 2020 Best Demonstration Paper, and a co-author of the Datasets paper, an EMNLP 2021 Best Demonstration Paper. Wolf also co-authored DistilBERT (2019), a distilled version of BERT, and TransferTransfo (2018), a transfer learning approach for conversational agents.⁴⁸⁹¹

Teaching and writing

Wolf co-authored the O'Reilly book Natural Language Processing with Transformers and produced educational material including the Ultra-Scale Playbook. Wolf co-taught a tutorial on transfer learning in NLP at NAACL 2019 and a tutorial on neural language generation at EMNLP 2020.¹

Work and contributions

• Transformers

  Listed on Wolf's site as a library Wolf created. Wolf made the repository's initial commit in October 2018 and is first author of the Transformers paper, which the README asks users to cite. The README's quickstart loads a model through the Pipeline API, which downloads and caches it for reuse.¹³⁴⁵

• Accelerate

  Made the repository's initial commit in October 2020 and is listed among the authors in the README's citation entry; Wolf's site names Accelerate among the libraries Wolf created or advised.⁶¹⁰¹

• SmolLM3

  Listed among the authors of Hugging Face's July 2025 announcement of SmolLM3, a 3B-parameter model in the SmolLM series.⁷

• Datasets: A Community Library for Natural Language Processing Open-source code

  Co-author of the 2021 paper describing Hugging Face's Datasets library, which standardizes interfaces, versioning and documentation for NLP datasets; Wolf's site lists it as an EMNLP 2021 Best Demonstration Paper.⁸¹

• DistilBERT Published research

  Co-author of the 2019 paper proposing DistilBERT, a smaller general-purpose language model pretrained by distilling BERT.⁹¹

• FineWeb Open data

  Co-author of the paper introducing the FineWeb pretraining datasets.¹¹

Timeline

1. Oct 29, 2018Made the initial commit to the huggingface/transformers repository.³
2. Jun 2, 2019Co-taught a NAACL 2019 tutorial on transfer learning in natural language processing.¹
3. Oct 2, 2019Released the DistilBERT paper with Victor Sanh, Lysandre Debut and Julien Chaumond.⁹
4. Oct 9, 2019Released the paper describing the Transformers library as first author.⁴
5. Oct 30, 2020Made the initial commit to the huggingface/accelerate repository.⁶
6. Nov 20, 2020Co-taught an EMNLP 2020 tutorial on neural language generation.¹
7. Sep 7, 2021Released the paper describing the Datasets library with Hugging Face colleagues.⁸
8. Jul 8, 2025Co-authored Hugging Face's announcement of the SmolLM3 model.⁷

Other documented roles

Roles the sources document that are not confirmed as current.

• Patent attorney (about five years, before Hugging Face), Cabinet Plasseraud¹

Honors and recognition

• EMNLP Best Demonstration Paper (Transformers) · Conference on Empirical Methods in Natural Language Processing (EMNLP), 2020¹
• EMNLP Best Demonstration Paper (Datasets) · Conference on Empirical Methods in Natural Language Processing (EMNLP), 2021¹

Where they work

• Hugging Face 34 people in the wiki

Related people

• Lewis TunstallAlso at Hugging Face
• Nils ReimersAlso at Hugging Face
• Lysandre DebutAlso at Hugging Face
• Margaret MitchellAlso at Hugging Face
• Anton LozhkovAlso at Hugging Face
• John GiannandreaLanguage models · ML frameworks and libraries
• Ronan CollobertML frameworks and libraries · Language models
• Piero MolinoML frameworks and libraries · Language models
• Samy BengioML frameworks and libraries · Language models
• Laurent SartranLanguage models · ML frameworks and libraries