FUSE-ing Language Models: Zero-Shot Adapter Discovery for Prompt Optimization Across Tokenizers
This addresses a domain-specific issue for researchers and practitioners working with multiple language models, but it is incremental as it builds on existing adapter and optimization techniques.
The paper tackled the problem of knowledge transfer across large language models with different tokenizers by proposing FUSE, an inexpensive method to approximate adapter layers for mapping embedding spaces, and demonstrated its efficacy in multi-objective optimization tasks like image captioning.
The widespread use of large language models has resulted in a multitude of tokenizers and embedding spaces, making knowledge transfer in prompt discovery tasks difficult. In this work, we propose FUSE (Flexible Unification of Semantic Embeddings), an inexpensive approach to approximating an adapter layer that maps from one model's textual embedding space to another, even across different tokenizers. We introduce a third-order tensor-based representation of a model's embedding space that aligns semantic embeddings that have been split apart by different tokenizers, and use this representation to derive an approximation of the gradient of one model's outputs with respect to another model's embedding space. We show the efficacy of our approach via multi-objective optimization over vision-language and causal language models for image captioning and sentiment-based image captioning.