Mechanistic Interpretability: Investigation of Large Language Models Steering by Modifying Internal Activations
Organisatsiooni nimi
Tartu NLP
Kokkuvõte
There is a line of work where we can find a special vector in Transformer model. This work digs into the nature of these vectors.
Lõputöö kaitsmise aasta
2024-2025
Juhendaja
Maksym Del
Suhtlemiskeel(ed)
inglise keel
Nõuded kandideerijale
You should understand and be able to tweak a Transformer architecture. The best fit is students who are good engineers aiming to possibly get a research publication.
Tase
Bakalaureus, Magister
Märksõnad
Kandideerimise kontakt
Nimi
Maksym Del
Tel
E-mail