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

Wanyun Xie, Francesco Tonin, Volkan Cevher
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:68725-68745, 2025.

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-xie25i, title = {Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning}, author = {Xie, Wanyun and Tonin, Francesco and Cevher, Volkan}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {68725--68745}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/xie25i/xie25i.pdf}, url = {https://proceedings.mlr.press/v267/xie25i.html}, 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.} }
Endnote
%0 Conference Paper %T Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning %A Wanyun Xie %A Francesco Tonin %A Volkan Cevher %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-xie25i %I PMLR %P 68725--68745 %U https://proceedings.mlr.press/v267/xie25i.html %V 267 %X 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.
APA
Xie, W., Tonin, F. & Cevher, V.. (2025). Chameleon: A Flexible Data-mixing Framework for Language Model Pretraining and Finetuning. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:68725-68745 Available from https://proceedings.mlr.press/v267/xie25i.html.

Related Material