PeLLE: Encoder-based language models for Brazilian Portuguese based on open data
This work addresses the need for specialized language models for Brazilian Portuguese users, but it is incremental as it adapts existing architectures to a new language with curated data.
The authors tackled the problem of lacking large language models for Brazilian Portuguese by introducing PeLLE, a family of RoBERTa-based models trained on curated open data, and found that while larger models generally perform better in downstream tasks, some tasks benefit from smaller models with curated pretraining data.
In this paper we present PeLLE, a family of large language models based on the RoBERTa architecture, for Brazilian Portuguese, trained on curated, open data from the Carolina corpus. Aiming at reproducible results, we describe details of the pretraining of the models. We also evaluate PeLLE models against a set of existing multilingual and PT-BR refined pretrained Transformer-based LLM encoders, contrasting performance of large versus smaller-but-curated pretrained models in several downstream tasks. We conclude that several tasks perform better with larger models, but some tasks benefit from smaller-but-curated data in its pretraining.