Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning

TL;DR

Chameleon is a flexible, efficient data-mixing framework for language model pretraining and finetuning that uses leverage scores to optimize domain importance, improving performance and adaptability without costly retraining.

Contribution

We introduce Chameleon, a novel data mixing method leveraging domain embeddings and leverage scores for efficient, adaptable domain reweighting in language model training.

Findings

Improves pretraining performance with less compute.

Enhances few-shot reasoning accuracy without retraining.

Consistently boosts test perplexity in finetuning scenarios.

Abstract

Training data mixtures greatly impact the generalization performance of large language models. Existing domain reweighting methods often rely on costly weight computations and require retraining when new data is introduced. To this end, we introduce a flexible and efficient data mixing framework, Chameleon, that employs leverage scores to quantify domain importance within a learned embedding space. We first construct a domain affinity matrix over domain embeddings. The induced leverage scores determine a mixture that upweights domains sharing common representations in embedding space. This formulation allows direct transfer to new data by computing the new domain embeddings. In experiments, we demonstrate improvements over three key scenarios: (i) our computed weights improve performance on pretraining domains with a fraction of the compute of existing methods; (ii) Chameleon can adapt to data changes without proxy retraining, boosting few-shot reasoning accuracies when transferred to new data; (iii) our method enables efficient domain reweighting in finetuning, consistently improving test perplexity on all finetuning domains over uniform mixture. Our code is available at https://github.com/LIONS-EPFL/Chameleon.