Dual-objective Language Models: Training Efficiency Without Overfitting
This incremental improvement addresses training efficiency and overfitting issues for language model developers, offering a flexible approach without architectural changes.
The paper tackles the trade-off between training efficiency and overfitting in language models by combining autoregressive and masked-diffusion objectives, showing that dual-objective training outperforms single-objective models across 50 evaluated settings with optimal balance consistency.
This paper combines autoregressive and masked-diffusion training objectives without any architectural modifications, resulting in flexible language models that outperform single-objective models. Autoregressive modeling has been a popular approach, partly because of its training efficiency; however, that comes at the cost of sensitivity to overfitting. On the other hand, masked-diffusion models are less efficient to train while being more resilient to overfitting. In this work, we demonstrate that dual-objective training achieves the best of both worlds. To derive the optimal balance between both objectives, we train and evaluate 50 language models under varying levels of data repetition. We show that it is optimal to combine both objectives under all evaluated settings and that the optimal balance is similar whether targeting autoregressive or masked-diffusion downstream performance.