Encoder-Decoder Gemma: Improving the Quality-Efficiency Trade-Off via Adaptation
This addresses the need for more efficient and effective models in real-world applications, though it is incremental as it builds on existing adaptation techniques.
The paper tackles the problem of adapting pretrained decoder-only large language models to encoder-decoder architectures to improve the trade-off between quality and efficiency, achieving results such as Gemma 2B-2B outperforming Gemma 2B by ~7% after instruction tuning.
While decoder-only large language models (LLMs) have shown impressive results, encoder-decoder models are still widely adopted in real-world applications for their inference efficiency and richer encoder representation. In this paper, we study a novel problem: adapting pretrained decoder-only LLMs to encoder-decoder, with the goal of leveraging the strengths of both approaches to achieve a more favorable quality-efficiency trade-off. We argue that adaptation not only enables inheriting the capability of decoder-only LLMs but also reduces the demand for computation compared to pretraining from scratch. We rigorously explore different pretraining objectives and parameter initialization/optimization techniques. Through extensive experiments based on Gemma 2 (2B and 9B) and a suite of newly pretrained mT5-sized models (up to 1.6B), we demonstrate the effectiveness of adaptation and the advantage of encoder-decoder LLMs. Under similar inference budget, encoder-decoder LLMs achieve comparable (often better) pretraining performance but substantially better finetuning performance than their decoder-only counterpart. For example, Gemma 2B-2B outperforms Gemma 2B by $\sim$7\% after instruction tuning. Encoder-decoder adaptation also allows for flexible combination of different-sized models, where Gemma 9B-2B significantly surpasses Gemma 2B-2B by $>$3\%. The adapted encoder representation also yields better results on SuperGLUE. We will release our checkpoints to facilitate future research.