Enabling Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration

Large language models (LLMs) have shown complementary strengths in various tasks and instances, motivating the research of ensembling LLMs to push the frontier leveraging the wisdom of the crowd. Existing work achieves this objective via training the extra reward model or fusion model to select or fuse all candidate answers. However, these methods pose a great challenge to the generalizability of the trained models. Besides, existing methods use the textual responses as communication media, ignoring the rich information in the inner representations of neural networks. Therefore, we propose a training-free ensemble framework DEEPEN, averaging the probability distributions outputted by different LLMs. A key challenge in this paradigm is the vocabulary discrepancy between heterogeneous LLMs, which hinders the operation of probability distribution averaging. To address this challenge, DEEPEN maps the probability distribution of each model from the probability space to a universe relative space based on the relative representation theory, and performs aggregation. Then, the result of aggregation is mapped back to the probability space of one LLM via a search-based inverse transformation to determine the generated token. We conduct experiments on the ensemble of various LLMs of 6B to 70B. Experimental results show that DEEPEN achieves consistent improvements across six popular benchmarks involving subject examination, reasoning and knowledge-QA, proving the effectiveness of our approach.

Aligning Translation-Specific Understanding to General Understanding in Large Language Models

Although large language models (LLMs) have shown surprising language understanding and generation capabilities, they have yet to gain a revolutionary advancement in the field of machine translation. One potential cause of the limited performance is the misalignment between the translation-specific understanding and general understanding inside LLMs. To align the translation-specific understanding to the general one, we propose a novel translation process xIoD (Cross-Lingual Interpretation of Difficult words), explicitly incorporating the general understanding on the content incurring inconsistent understanding to guide the translation. Specifically, xIoD performs the cross-lingual interpretation for the difficult-to-translate words and enhances the translation with the generated interpretations. Furthermore, we reframe the external tools of QE to tackle the challenges of xIoD in the detection of difficult words and the generation of helpful interpretations. We conduct experiments on the self-constructed benchmark ChallengeMT, which includes cases in which multiple SOTA translation systems consistently underperform. Experimental results show the effectiveness of our xIoD, which improves up to +3.85 COMET.

Towards Higher Pareto Frontier in Multilingual Machine Translation

Multilingual neural machine translation has witnessed remarkable progress in recent years. However, the long-tailed distribution of multilingual corpora poses a challenge of Pareto optimization, i.e., optimizing for some languages may come at the cost of degrading the performance of others. Existing balancing training strategies are equivalent to a series of Pareto optimal solutions, which trade off on a Pareto frontier. In this work, we propose a new training framework, Pareto Mutual Distillation (Pareto-MD), towards pushing the Pareto frontier outwards rather than making trade-offs. Specifically, Pareto-MD collaboratively trains two Pareto optimal solutions that favor different languages and allows them to learn from the strengths of each other via knowledge distillation. Furthermore, we introduce a novel strategy to enable stronger communication between Pareto optimal solutions and broaden the applicability of our approach. Experimental results on the widely-used WMT and TED datasets show that our method significantly pushes the Pareto frontier and outperforms baselines by up to +2.46 BLEU.