LLMR: Knowledge Distillation with a Large Language Model-Induced Reward
This work addresses the deployment challenge of large language models for NLP applications, but it appears incremental as it builds on existing knowledge distillation techniques.
The paper tackles the problem of deploying large language models in resource-constrained environments by proposing LLMR, a knowledge distillation method using a reward function induced from large language models, and it shows that LLMR consistently outperforms traditional knowledge distillation methods on multiple datasets in dialogue generation and summarization tasks.
Large language models have become increasingly popular and demonstrated remarkable performance in various natural language processing (NLP) tasks. However, these models are typically computationally expensive and difficult to be deployed in resource-constrained environments. In this paper, we propose LLMR, a novel knowledge distillation (KD) method based on a reward function induced from large language models. We conducted experiments on multiple datasets in the dialogue generation and summarization tasks. Empirical results demonstrate that our LLMR approach consistently outperforms traditional KD methods in different tasks and datasets.