Lewis Tunstall

Open-source work

Lewis Tunstall is a machine learning engineer at Hugging Face who works on post-training and open-source tools for the wider community. Tunstall's GitHub profile highlights work on TRL, Hugging Face's post-training library, the Alignment Handbook of training recipes, and Open-R1.²¹⁵⁴

In January 2025 Tunstall co-authored the Hugging Face blog post launching Open-R1, a project to build an open reproduction of the DeepSeek-R1 reasoning pipeline. Tunstall is listed first in the Alignment Handbook's citation entry and among the authors of TRL.⁶⁴⁵

Research

Tunstall is first author of "Efficient Few-Shot Learning Without Prompts" (2022), which introduced SetFit, a method that fine-tunes Sentence Transformers contrastively on a small number of labeled examples. Tunstall is also first author of the Zephyr paper (2023), which aligned a 7B chat model with distilled direct preference optimization on AI-feedback data.⁷³

Teaching and writing

Tunstall co-wrote the O'Reilly book Natural Language Processing with Transformers, whose co-authors include Leandro von Werra. Tunstall is also one of the authors of Hugging Face's LLM Course.²

Work and contributions

• TRL (Transformers Reinforcement Learning) Open-source code

  Listed among the authors in TRL's citation file. TRL is an Apache-2.0 toolkit for post-training transformer models, with implementations of SFT, reward modeling, DPO, and GRPO.⁵

• Zephyr Published research

  First author of "Zephyr: Direct Distillation of LM Alignment" (2023), which trained a smaller chat model with distilled supervised fine-tuning and distilled direct preference optimization on preference data ranked by a teacher model.³

• The Alignment Handbook Open-source code

  Listed first in the citation entry of the repository, which provides training recipes for continued pretraining and for aligning language models with human and AI preferences.⁴

• Natural Language Processing with Transformers (O'Reilly) Published research

  Co-author of the O'Reilly book, according to the author biographies on Hugging Face's LLM Course page.²

Timeline

1. Sep 2022Published "Efficient Few-Shot Learning Without Prompts", introducing SetFit, as first author.⁷
2. Oct 2023Published the Zephyr paper on direct distillation of language model alignment as first author.³
3. Jan 28, 2025Co-authored the Hugging Face blog post launching the Open-R1 project.⁶

Where they work

• Hugging Face 34 people in the wiki

Related people

• Lysandre DebutAlso at Hugging Face
• Leandro von WerraAlso at Hugging Face
• Younes BelkadaAlso at Hugging Face
• Nils ReimersAlso at Hugging Face
• Edward BeechingAlso at Hugging Face
• Alexander WettigLanguage models · Reinforcement learning
• Albert JiangLanguage models · Reinforcement learning
• Nathan LambertLanguage models · Reinforcement learning
• Jonathan CohenLanguage models · ML frameworks and libraries
• Mika SenghaasReinforcement learning · Language models