Learning to Route Among Specialized Experts for Zero-Shot Generalization
Mohammed Muqeeth, Haokun Liu, Yufan Liu, Colin Raffel
TL;DR
PHATGOOSE introduces a post-hoc, tokenwise gating mechanism to route among independently trained specialized PEFT modules, enabling zero-shot generalization without sharing datasets. By training lightweight gates after freezing both the base model and the expert modules, and performing per-layer, per-token top-k routing, PHATGOOSE combines diverse expert capabilities to improve unseen task performance. Across T5-family experiments with two expert pools, PHATGOOSE consistently outperforms prior post-hoc baselines and can rival explicit multitask training, demonstrating the value of decentralized, adaptive expert recycling. The work also provides qualitative insight into diverse routing strategies and sets the stage for expanding decentralized collaboration in generalist AI development.
Abstract
Recently, there has been a widespread proliferation of "expert" language models that are specialized to a specific task or domain through parameter-efficient fine-tuning. How can we recycle large collections of expert language models to improve zero-shot generalization to unseen tasks? In this work, we propose Post-Hoc Adaptive Tokenwise Gating Over an Ocean of Specialized Experts (PHATGOOSE), which learns to route among specialized modules that were produced through parameter-efficient fine-tuning. Unlike past methods that learn to route among specialized models, PHATGOOSE explores the possibility that zero-shot generalization will be improved if different experts can be adaptively chosen for each token and at each layer in the model. Crucially, our method is post-hoc - it does not require simultaneous access to the datasets used to create the specialized models and only requires a modest amount of additional compute after each expert model is trained. In experiments covering a range of specialized model collections and zero-shot generalization benchmarks, we find that PHATGOOSE outperforms past methods for post-hoc routing and, in some cases, outperforms explicit multitask training (which requires simultaneous data access). To better understand the routing strategy learned by PHATGOOSE, we perform qualitative experiments to validate that PHATGOOSE's performance stems from its ability to make adaptive per-token and per-module expert choices. We release all of our code to support future work on improving zero-shot generalization by recycling specialized experts.