5 papers
On the Emergence of Induction Heads for In-Context Learning
Tiberiu Musat, Tiago Pimentel, Lorenzo Noci +3
Transformers have become the dominant architecture for natural language processing. Part of their success is owed to a remarkable capability known as in-context learning (ICL): the…
Convergence and Divergence of Language Models under Different Random Seeds
Finlay Fehlauer, Kyle Mahowald, Tiago Pimentel
In this paper, we investigate the convergence of language models (LMs) trained under different random seeds, measuring convergence as the expected per-token Kullback--Leibler (KL)…
The Non-Linear Representation Dilemma: Is Causal Abstraction Enough for Mechanistic Interpretability?
Denis Sutter, Julian Minder, Thomas Hofmann +1
The concept of causal abstraction got recently popularised to demystify the opaque decision-making processes of machine learning models; in short, a neural network can be abstracte…
Causal Estimation of Tokenisation Bias
Pietro Lesci, Clara Meister, Thomas Hofmann +2
Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser -- which maps…
Local and Global Decoding in Text Generation
Daniel Gareev, Thomas Hofmann, Ezhilmathi Krishnasamy +1
Text generation, a key component in applications such as dialogue systems, relies on decoding algorithms that sample strings from a language model distribution. Traditional methods…