Continual Model Routing in Evolving Model Hubs

Jack Bell, Giacomo Carfi’, Gerlando Gramaglia, Vincenzo Lomonaco
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:7412-7453, 2026.

Abstract

AI model hubs provide access to a rapidly growing collection of powerful pre-trained models, enabling off-the-shelf mixture-of-experts systems with different routing strategies. However, this rapid growth poses two fundamental challenges: scaling model selection across thousands of experts and continually updating routing mechanisms as new models and tasks are introduced. In this paper, we formalise this setting as Continual Model Routing (CMR) and propose CMRBench, a new large-scale benchmark simulating realistic hub expansion and including over 2,000 candidate models. Finally, we introduce CARvE, a contrastive embedding approach for efficient continual model routing via domain-stratified coreset replay and checkpoint-based anchoring. Extensive empirical results and ablations show that CARvE significantly outperforms zero-shot retrieval, fine-tuning, and adapter-merging baselines in model, family, and domain-level accuracy.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-bell26a, title = {Continual Model Routing in Evolving Model Hubs}, author = {Bell, Jack and Carfi', Giacomo and Gramaglia, Gerlando and Lomonaco, Vincenzo}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {7412--7453}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/bell26a/bell26a.pdf}, url = {https://proceedings.mlr.press/v306/bell26a.html}, abstract = {AI model hubs provide access to a rapidly growing collection of powerful pre-trained models, enabling off-the-shelf mixture-of-experts systems with different routing strategies. However, this rapid growth poses two fundamental challenges: scaling model selection across thousands of experts and continually updating routing mechanisms as new models and tasks are introduced. In this paper, we formalise this setting as Continual Model Routing (CMR) and propose CMRBench, a new large-scale benchmark simulating realistic hub expansion and including over 2,000 candidate models. Finally, we introduce CARvE, a contrastive embedding approach for efficient continual model routing via domain-stratified coreset replay and checkpoint-based anchoring. Extensive empirical results and ablations show that CARvE significantly outperforms zero-shot retrieval, fine-tuning, and adapter-merging baselines in model, family, and domain-level accuracy.} }
Endnote
%0 Conference Paper %T Continual Model Routing in Evolving Model Hubs %A Jack Bell %A Giacomo Carfi’ %A Gerlando Gramaglia %A Vincenzo Lomonaco %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-bell26a %I PMLR %P 7412--7453 %U https://proceedings.mlr.press/v306/bell26a.html %V 306 %X AI model hubs provide access to a rapidly growing collection of powerful pre-trained models, enabling off-the-shelf mixture-of-experts systems with different routing strategies. However, this rapid growth poses two fundamental challenges: scaling model selection across thousands of experts and continually updating routing mechanisms as new models and tasks are introduced. In this paper, we formalise this setting as Continual Model Routing (CMR) and propose CMRBench, a new large-scale benchmark simulating realistic hub expansion and including over 2,000 candidate models. Finally, we introduce CARvE, a contrastive embedding approach for efficient continual model routing via domain-stratified coreset replay and checkpoint-based anchoring. Extensive empirical results and ablations show that CARvE significantly outperforms zero-shot retrieval, fine-tuning, and adapter-merging baselines in model, family, and domain-level accuracy.
APA
Bell, J., Carfi’, G., Gramaglia, G. & Lomonaco, V.. (2026). Continual Model Routing in Evolving Model Hubs. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:7412-7453 Available from https://proceedings.mlr.press/v306/bell26a.html.

Related Material