[B]. In..."> [B]. In..."> [B]. In...">

In-context Learning and Induction Heads

by   Catherine Olsson, et al.

"Induction heads" are attention heads that implement a simple algorithm to complete token sequences like [A][B] ... [A] -> [B]. In this work, we present preliminary and indirect evidence for a hypothesis that induction heads might constitute the mechanism for the majority of all "in-context learning" in large transformer models (i.e. decreasing loss at increasing token indices). We find that induction heads develop at precisely the same point as a sudden sharp increase in in-context learning ability, visible as a bump in the training loss. We present six complementary lines of evidence, arguing that induction heads may be the mechanistic source of general in-context learning in transformer models of any size. For small attention-only models, we present strong, causal evidence; for larger models with MLPs, we present correlational evidence.


page 10

page 21

page 30

page 32

page 34

page 36

page 37

page 38


Rethinking the Role of Scale for In-Context Learning: An Interpretability-based Case Study at 66 Billion Scale

Language models have been shown to perform better with an increase in sc...

Induction and Skolemization in saturation theorem proving

We consider a typical integration of induction in saturation-based theor...

CodeFill: Multi-token Code Completion by Jointly Learning from Structure and Naming Sequences

Code completion is an essential feature of IDEs, yet current autocomplet...

Induction Models on ℕ

Mathematical induction is a fundamental tool in computer science and mat...

Positivity certificates for linear recurrences

We show that for solutions of linear recurrences with polynomial coeffic...

A Study of the Attention Abnormality in Trojaned BERTs

Trojan attacks raise serious security concerns. In this paper, we invest...

Inducing a Semantically Annotated Lexicon via EM-Based Clustering

We present a technique for automatic induction of slot annotations for s...

Please sign up or login with your details

Forgot password? Click here to reset