PhoB

PhoB is a family of pre-trained language models for Vietnamese developed by VinAI Research. Built on the Transformer encoder architecture, PhoB models aim to provide language-specific representations that improve natural language understanding for Vietnamese compared with multilingual models trained on mixed languages.

The core of the PhoB family is PhoBERT, which includes base and larger configurations. These models are pre-trained on large-scale Vietnamese corpora collected from diverse sources using a subword tokenization scheme and a masked language modeling objective to learn contextual representations. They are released with pretrained weights and tooling compatible with common deep learning frameworks, making them accessible for research and development.

PhoB models are intended to serve as strong starting points for fine-tuning on downstream Vietnamese NLP tasks. They have been evaluated on a range of tasks, including sentiment analysis, named entity recognition, part-of-speech tagging, and text classification, where they typically achieve strong performance relative to baseline models and multilingual alternatives. The approach has contributed to improvements in both academic research and practical deployments for Vietnamese language processing.

The project is maintained by VinAI Research and collaborators, with code and pretrained weights widely used in the Vietnamese NLP community. PhoB models are commonly adopted to accelerate development of domain-specific applications, enabling researchers and developers to build higher-quality Vietnamese NLP systems. See also related Vietnamese NLP resources and pre-trained models in the broader language-model ecosystem.