Features have life history. And we should care

TL;DR

This paper investigates the life history of features in language models, revealing a stable, load-bearing core that emerges early and guides subsequent development, with implications for understanding model training dynamics.

Contribution

It identifies a persistent representational backbone in language models, characterizes its properties, and shows its early emergence and influence on training.

Findings

A stable scaffold of ~50 features exists in language models.

The scaffold emerges rapidly within the first 1% of training.

Scaffold features are highly load-bearing and predictive of future carriers.

Abstract

Features in language models have life history: they emerge, persist, and die during training, yet the importance of that history remains largely unexplored. We find evidence of a persistent representational backbone, which we identify in Pythia-160M and -410M as the carrier scaffold: ${\sim}50$ sparse features with stable life histories, around which the model's representational structure organises. It has four properties. \emph{(i)}~\emph{It assembles early:} features emerge, die, and reorganise ${\sim}40\!\times$ faster in the first $1\%$ of training than afterwards, and the scaffold is already largely fixed by then. \emph{(ii)}~\emph{It is load-bearing:} joint cross-layer ablation identifies the carriers as far more load-bearing than any count-matched non-scaffold population, a gap invisible to per-firing single-feature methods. \emph{(iii)}~\emph{Function precedes direction:} which features will become carriers is already predictable from training-onset firing patterns alone, correctly distinguishing future carriers from non-carriers in $4$ of $5$ cases, before the geometry has settled. \emph{(iv)}~\emph{It seeds subsequent development:} by the end of training, scaffold carriers have recruited $64\%$ of all active features into the scaffold hierarchy. Life history is consistent with a two-phase account of training: selection appears to largely determine the scaffold in the first $1\%$; the remaining $99\%$ appears to calibrate geometry around a substrate already set.