Search papers, labs, and topics across Lattice.
This study systematically investigates the selection of student layers for zero-shot model size interpolation, framing the optimization of layer subsets as a shortest-path problem in an acyclic graph. The authors demonstrate that layer patching significantly influences interpolation behavior, with performance varying across different model families. Their greedy algorithm, KLPatch, based on KL divergence, outperforms traditional sequential patching methods, providing a robust framework for constructing effective interpolated models.
Layer patching can dramatically enhance model performance in size interpolation, revealing that simple strategies often outperform complex methods.
Zero-shot model size interpolation aims to create new models of intermediate target sizes by combining existing models without additional training. Recent work on boomerang distillation [Kangaslahti et al., 2026] shows that a student language model distilled from a larger teacher can be expanded by iteratively patching its layers, replacing student layers with contiguous blocks of teacher layers to obtain models whose size and performance interpolate between the student and the teacher. In this work, we provide the first systematic study of student-layer selection for model size interpolation. We cast finding the optimal layer subset for each model size as an optimization problem and prove it can be viewed as a shortest-path problem in a certain acyclic graph. In experiments, we show that patching strongly shapes interpolation behavior, with effects that vary substantially across model families. We find that simple sequential strategies--patching either from the first layer to the last or from the last to the first--often achieve surprisingly strong performance in practice. We further introduce KLPatch, a greedy patching algorithm based on KL divergence, which often improves over last-to-first patching and approximately solves the optimization problem. Together, our results provide a principled understanding of how layer patching affects model size interpolation and offer practical guidance for constructing near-optimal interpolated models.
Microsoft Research