Lewis Tunstall: Hugging Face, SetFit and Reinforcement Learning | Learning from Machine Learning #6

This episode features Lewis Tunstall, machine learning engineer at Hugging Face and author of the best selling book Natural Language Processing with Transformers. He currently focuses on one of the hottest topic in NLP right now reinforcement learning from human feedback (RLHF). Lewis holds a PhD in quantum physics and his research has taken him around the world and into some of the most impactful projects including the Large Hadron Collider, the world's largest and most powerful particle accelerator. Lewis shares his unique story from Quantum Physicist to Data Scientist to Machine Learning Engineer.

Fun Facts and Takeaways

1️⃣ Lewis started his Machine Learning journey after seeing 100 lines of TensorFlow code crush benchmarks particle physics researchers had been working on for about 20 years.
2️⃣ Coming from a physics background, combined with his first coding experience being with Fortran, he learned Python later in his career.
3️⃣ Taking fast.ai courses helped him adopt a more practical approach - learning by doing and solving specific problems instead of trying to master languages and tools first.
4️⃣ Contributing to open source documentation and libraries was hugely valuable, as it forced him to really understand the internals to be able to explain concepts to others.
5️⃣ He co-authored the popular book "Natural Language Processing with Transformers," to fill the gap between academic papers and conceptual overviews of transformers, before joining Hugging Face!
6️⃣ Lewis helped created SetFit, an efficient few-shot learning method, to help with industry problems of limited labeled data (and one of my favorite libraries)
7️⃣ He is currently working on scaling Reinforcement Learning for Human Preference Learning to make LLMs more aligned
8️⃣ Like many in the field, Lewis and Hugging Face give chatbots a “vibes test” - listen in for some of the funny scenarios!

Lewis is a kind and humble person, and I really appreciate him giving me the opportunity to discuss some of the hottest topics in Artificial Intelligence and Natural Language Processing. It was an absolute pleasure and I look forward to continuing to unpack the lessons learned from this episode.

---

Resources to learn more about Lewis Tunstall

• https://www.linkedin.com/in/lewis-tunstall/

• https://github.com/lewtun

---

References from the Episode

• Hugging Face

• Fast.ai

• Jeremy Howard

• SetFit - https://arxiv.org/abs/2209.11055

• Proximal Policy Optimization

• InstructGPT

• RAFT Benchmark

• Bidirectional Language Models are Also Few-Shot Learners

• Nils Reimers - Sentence Transformers

• Jay Alammar - Illustrated Transformer

• Annotated Transformer

• Moshe Wasserblat, Intel, NLP, Research Manager

• Leandro von Werra, Co-Author of NLP with Transformers, Hugging Face Researcher

• John Schulman

• LLMSys

• LoRA - Low-Rank Adaptation of Large Language Models

---

Resources to learn more about Learning from Machine Learning

• Learning from Machine Learning on LinkedIn

• Connect with me on LinkedIn

• Check out my articles on Medium

• Podcast on all platforms