Beyond Output Matching: Bidirectional Alignment for Enhanced In-Context Learning
This work addresses deployment challenges for large language models by enhancing the efficiency of smaller models, representing an incremental improvement over existing distillation methods.
The paper tackles the problem of transferring in-context learning abilities from large to compact language models by proposing Bidirectional Alignment, which aligns both input preferences and output distributions, resulting in consistent performance improvements across language understanding, reasoning, and coding tasks.
Large language models (LLMs) have shown impressive few-shot generalization on many tasks via in-context learning (ICL). Despite their success in showing such emergent abilities, the scale and complexity of larger models also lead to unprecedentedly high computational demands and deployment challenges. In reaction, researchers explore transferring the powerful capabilities of larger models to more efficient and compact models by typically aligning the output of smaller (student) models with that of larger (teacher) models. Existing methods either train student models on the generated outputs of teacher models or imitate their token-level probability distributions. However, these distillation methods pay little to no attention to the input, which also plays a crucial role in ICL. Based on the finding that the performance of ICL is highly sensitive to the selection of demonstration examples, we propose Bidirectional Alignment (BiAlign) to fully leverage the models' preferences for ICL examples to improve the ICL abilities of student models. Specifically, we introduce the alignment of input preferences between student and teacher models by incorporating a novel ranking loss, in addition to aligning the token-level output distribution. With extensive experiments and analysis, we demonstrate that BiAlign can consistently outperform existing baselines on a variety of tasks involving language understanding, reasoning, and coding.