UniLM is Microsoft's open-source research repository for unified language and multimodal model work. It collects model implementations, pretrained checkpoints, papers, and code spanning language understanding and generation, document intelligence, vision-language learning, speech-related tasks, and other areas developed across multiple research projects.

The repository has served as a home for influential projects such as UniLM, LayoutLM and related document models, BEiT-family research, multimodal systems, and various pretrained-model experiments. Individual subprojects typically include their own training, evaluation, and inference instructions rather than sharing one single application API.

UniLM is primarily a research collection for practitioners who want to reproduce Microsoft Research results or build on specific model families. Its breadth makes it valuable as an archive of pretrained-model techniques, though users generally work with one subproject at a time rather than installing the repository as a unified end-user product.