unilm

Why we included this project

Microsoft's unilm repository is the company's central research repository for pre-trained foundation models, and it is a good place to start if you want to understand or reuse the models behind document AI, multilingual NLP, and multimodal systems. It collects released checkpoints and training code: LayoutLM and TrOCR read scanned documents and PDFs, Kosmos grounds language in images, and MiniLM and E5 provide compact embeddings for retrieval. It is not one deployable product but a catalog of pretrained weights and recipes, each with its own subfolder, papers, and fine-tuning scripts. Teams building document understanding or cross-lingual search will find ready-made starting points here, while researchers get a clear view of how Microsoft's models evolved across tasks, languages, and modalities.