Masked Diffusion Language Models (MDLM) have recently demonstrated that discrete diffusion can achieve competitive performance in text generation. However, training these models remains computationally expensive, particularly for lower-resourced languages like Portuguese. In this work, we adapt REPresentation Alignment (REPA), a technique originally proposed for vision, to the textual domain. We systematically evaluate the impact of aligning the internal representations of a Portuguese MDLM with those of pretrained teacher encoders (e.g., Qwen, BERTimbau). Our experiments show that REPA significantly accelerates training and improves final perplexity by 28.6% compared to a baseline without alignment. We also identify optimal hyperparameters, finding that mid-level alignment with modern teacher encoders yields the best results.

In this paper, we delineate the strategy employed by our team, DeepLearningBrasil, which secured us the first place in the shared task DepSign-LT-EDI@RANLP-2023 with the advantage of 2.4%. The task was to classify social media texts into three distinct levels of depression - “not depressed,” “moderately depressed,” and “severely depressed.” Leveraging the power of the RoBERTa and DeBERTa models, we further pre-trained them on a collected Reddit dataset, specifically curated from mental health-related Reddit’s communities (Subreddits), leading to an enhanced understanding of nuanced mental health discourse. To address lengthy textual data, we introduced truncation techniques that retained the essence of the content by focusing on its beginnings and endings. Our model was robust against unbalanced data by incorporating sample weights into the loss. Cross-validation and ensemble techniques were then employed to combine our k-fold trained models, delivering an optimal solution. The accompanying code is made available for transparency and further development.