microsoft/unilm

Health v2 · maintenance only

Adoption not part of the score

22194 stars · 2702 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Microsoft's collection of large-scale self-supervised pre-trained models spanning tasks, 100+ languages, and modalities (text, image, layout, audio). It includes well-known models like UniLM, LayoutLM, BEiT, TroCR, Kosmos, and BitNet, plus transformer architecture research such as DeepNet and LongNet.

Use cases

• pretrain a transformer model on multiple languages
• extract text and layout from scanned documents
• run OCR-free document understanding
• fine-tune a multimodal large language model
• train a vision transformer for image classification
• experiment with 1-bit LLM architectures
• build a multilingual sentence embedding model

When to choose

• you need state-of-the-art pre-trained models for NLP, vision, or document AI
• you want to reproduce or build on published Microsoft research models
• you need multilingual or multimodal (vision+language) model checkpoints

When to avoid

• you need a production-ready turnkey API rather than research code
• you want a lightweight inference-only library without training pipelines
• you need a single unified model interface - the repo is a collection of separate subprojects

Facets

library · maturity active

machine-learning deep-learning nlp llm-training transformers computer-vision ocr machine-learning deep-learning large-language-models computer-vision artificial-intelligence python cross-platform pre-trained-models multimodal document-ai self-supervised-learning foundation-models research transformer-architectures natural-language-processing gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="microsoft/unilm")