Neurons in Large Language Models: Dead, N-gram, Positional

TL;DR

This study investigates large language models, revealing that many neurons are inactive or specialized for discrete features, with some neurons explicitly removing information, and positional neurons vary with model size.

Contribution

It provides a lightweight analysis method for large models, uncovering neuron sparsity, feature detection, and information removal mechanisms across different scales.

Findings

Over 70% of neurons can be 'dead' in large models.

Many neurons act as token and n-gram detectors.

Some neurons explicitly remove information from the residual stream.

Abstract

We analyze a family of large language models in such a lightweight manner that can be done on a single GPU. Specifically, we focus on the OPT family of models ranging from 125m to 66b parameters and rely only on whether an FFN neuron is activated or not. First, we find that the early part of the network is sparse and represents many discrete features. Here, many neurons (more than 70% in some layers of the 66b model) are "dead", i.e. they never activate on a large collection of diverse data. At the same time, many of the alive neurons are reserved for discrete features and act as token and n-gram detectors. Interestingly, their corresponding FFN updates not only promote next token candidates as could be expected, but also explicitly focus on removing the information about triggering them tokens, i.e., current input. To the best of our knowledge, this is the first example of mechanisms specialized at removing (rather than adding) information from the residual stream. With scale, models become more sparse in a sense that they have more dead neurons and token detectors. Finally, some neurons are positional: them being activated or not depends largely (or solely) on position and less so (or not at all) on textual data. We find that smaller models have sets of neurons acting as position range indicators while larger models operate in a less explicit manner.