Understanding Emergent Abilities of Language Models from the Loss Perspective

TL;DR

This paper investigates emergent abilities in language models through the lens of pre-training loss, revealing that such abilities emerge when models reach a specific loss threshold, independent of size or compute.

Contribution

It introduces a loss-based perspective to understand emergent abilities, showing they depend on pre-training loss rather than model size or training compute.

Findings

Models with the same pre-training loss perform similarly across tasks.

Emergent abilities appear when pre-training loss drops below a threshold.

Performance remains at random guessing levels above the threshold.

Abstract

Recent studies have put into question the belief that emergent abilities in language models are exclusive to large models. This skepticism arises from two observations: 1) smaller models can also exhibit high performance on emergent abilities and 2) there is doubt on the discontinuous metrics used to measure these abilities. In this paper, we propose to study emergent abilities in the lens of pre-training loss, instead of model size or training compute. We demonstrate that the Transformer models with the same pre-training loss, but different model and data sizes, generate the same performance on various downstream tasks, with a fixed data corpus, tokenization, and model architecture. We also discover that a model exhibits emergent abilities on certain tasks -- regardless of the continuity of metrics -- when its pre-training loss falls below a specific threshold. Before reaching this threshold, its performance remains at the level of random guessing. This inspires us to redefine emergent abilities as those that manifest in models with lower pre-training losses, highlighting that these abilities cannot be predicted by merely extrapolating the performance trends of models with higher pre-training losses.