The Finetuner's Fallacy: When to Pretrain with Your Finetuning Data

TL;DR

This paper introduces specialized pretraining (SPT), a method that improves domain-specific performance and preserves general knowledge by repeating small domain datasets during pretraining, reducing overfitting and compute needs.

Contribution

The paper proposes SPT, a novel pretraining strategy that enhances domain adaptation and knowledge retention compared to standard pretraining methods.

Findings

SPT improves domain performance and preserves general capabilities.

SPT reduces pretraining tokens needed by up to 1.75x.

SPT outperforms larger models on underrepresented domains.

Abstract

Real-world model deployments demand strong performance on narrow domains where data is often scarce. Typically, practitioners finetune models to specialize them, but this risks overfitting to the domain and forgetting general knowledge. We study a simple strategy, specialized pretraining (SPT), where a small domain dataset, typically reserved for finetuning, is repeated starting from pretraining as a fraction of the total tokens. Across three specialized domains (ChemPile, MusicPile, and ProofPile), SPT improves domain performance and preserves general capabilities after finetuning compared to standard pretraining. In our experiments, SPT reduces the pretraining tokens needed to reach a given domain performance by up to 1.75x. These gains grow when the target domain is underrepresented in the pretraining corpus: on domains far from web text, a 1B SPT model outperforms a 3B standard pretrained model. Beyond these empirical gains, we derive overfitting scaling laws to guide practitioners in selecting the optimal domain-data repetition for a given pretraining compute budget. Our observations reveal the finetuner's fallacy: while finetuning may appear to be the cheapest path to domain adaptation, introducing specialized domain data during pretraining stretches its utility. SPT yields better specialized domain performance (via reduced overfitting across repeated exposures) and better general domain performance (via reduced forgetting during finetuning), ultimately achieving stronger results with fewer parameters and less total compute when amortized over inference. To get the most out of domain data, incorporate it as early in training as possible.