Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla

TL;DR

This paper investigates whether circuit analysis techniques can scale to large language models like Chinchilla 70B, demonstrating their effectiveness in understanding multiple-choice question answering mechanisms and identifying key output nodes.

Contribution

It extends circuit analysis methods to a large-scale model, showing their scalability and providing insights into the model's handling of multiple-choice questions and answer label representations.

Findings

Circuit analysis techniques scale to Chinchilla 70B.

Output nodes like attention heads and MLPs can be identified and categorized.

Query and key subspaces encode enumeration features.

Abstract

\emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \emph{label} given knowledge of the correct answer \emph{text}. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.