ZurichNLP #9
Lewis Tunstall from Hugging Face on LLM alignment and Vilém Zouhar from ETH Zurich on tokenization.
Loading…
182 registered
Speakers
Lewis Tunstall
Hugging FaceVilém Zouhar
ETH ZurichAbout
We’re happy to announce ZurichNLP #9 on April 25th with the following speakers:
Lewis Tunstall (MLE @ HuggingFace) on How to Align Your LLM
Vilém Zouhar (PhD @ ETH Zurich) on Pride and BPE: How We Solved Tokenization but Got It Wrong
“Tokenization is present in almost all NLP pipelines, but rarely examined mathematically. We formalize and show boundaries to the most popular tokenization algorithm, Byte-Pair Encoding. Then, with information theory, we show what makes some tokenization better than others and how to use this as a metric before training your expensive models. Lastly, we admit how we got this hypothesis wrong.”
Slides from Lewis’s presentation can be found here!
RSVP soon as spots are limited.
Programme
- How to Align Your LLMLewis Tunstall
- Pride and BPE: How We Solved Tokenization but Got It WrongVilém Zouhar