MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods
This work addresses the efficiency bottleneck in NLG decoding for researchers and practitioners, offering a way to achieve state-of-the-art performance without expensive inference, though it is incremental as it builds on existing decoding methods.
The paper tackles the high computational cost of advanced decoding methods like Minimum Bayes' Risk (MBR) and Quality Estimation (QE) reranking in Natural Language Generation by proposing MBR and QE finetuning to distill their quality gains into models during training, enabling efficient inference; results show these methods significantly outperform base models and even surpass finetuning on human-generated references when using an external LLM as a teacher.
Recent research in decoding methods for Natural Language Generation (NLG) tasks has shown that MAP decoding is not optimal, because model probabilities do not always align with human preferences. Stronger decoding methods, including Quality Estimation (QE) reranking and Minimum Bayes' Risk (MBR) decoding, have since been proposed to mitigate the model-perplexity-vs-quality mismatch. While these decoding methods achieve state-of-the-art performance, they are prohibitively expensive to compute. In this work, we propose MBR finetuning and QE finetuning which distill the quality gains from these decoding methods at training time, while using an efficient decoding algorithm at inference time. Using the canonical NLG task of Neural Machine Translation (NMT), we show that even with self-training, these finetuning methods significantly outperform the base model. Moreover, when using an external LLM as a teacher model, these finetuning methods outperform finetuning on human-generated references. These findings suggest new ways to leverage monolingual data to achieve improvements in model quality that are on par with, or even exceed, improvements from human-curated data, while maintaining maximum efficiency during decoding.