Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning
This work addresses the challenge of efficient and flexible data mixing for language model pretraining and finetuning, offering incremental improvements over existing methods.
The paper tackles the problem of optimizing data mixtures for language model training by introducing Chameleon, a framework that uses leverage scores to quantify domain importance, which improves performance on pretraining domains with reduced compute and adapts to new data without retraining.
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.