Transporting Task Vectors across Different Architectures without Training
This work addresses the challenge of expensive relearning for adapting pre-trained models to downstream tasks, offering a novel solution for cross-architecture transfer, though it is incremental in extending transfer beyond identical architectures.
The paper tackled the problem of transferring task-specific parameter updates across models with different architectures without retraining, and introduced Theseus, a training-free method that achieves consistent improvements over strong baselines in vision and language models.
Adapting large pre-trained models to downstream tasks often produces task-specific parameter updates that are expensive to relearn for every model variant. While recent work has shown that such updates can be transferred between models with identical architectures, transferring them across models of different widths remains largely unexplored. In this work, we introduce Theseus, a training-free method for transporting task-specific updates across heterogeneous models. Rather than matching parameters directly, we characterize a task update by the functional effect it induces on intermediate representations. We formalize task-vector transport as a functional matching problem on observed activations and show that, after aligning representation spaces via orthogonal Procrustes analysis, it admits a stable closed-form solution that preserves the geometry of the update. We evaluate Theseus on vision and language models across different widths, showing consistent improvements over strong baselines without additional training or backpropagation. Our results show that task updates can be meaningfully transferred across architectures when task identity is defined functionally rather than parametrically.