Extension Modules

All examples target MinerU 4.x and run inside an activated virtual environment.

Installation	Includes	Use case
mineru>=4.0,<5	ONNX, llama.cpp, WebUI, SDK/API; Torch is automatic on Apple Silicon	Native documents, CPU small models, llama.cpp, or an existing service
mineru[torch]>=4.0,<5	Base package + Torch, Torchvision, Transformers, Accelerate	Torch small models with an independently selected VLM engine
mineru[full]>=4.0,<5	Torch extra + vLLM on Linux / LMDeploy on Windows	Higher-throughput serving on supported devices

uv pip install -U "mineru[torch]>=4.0,<5"

Or install the platform-specific inference engine:

uv pip install -U "mineru[full]>=4.0,<5"

all composes full. core, pipeline, vlm, vllm, lmdeploy, and gradio are no longer MinerU 4.0 extras. Tier names are not extras either.

Optional engine constraints

The current 4.0 dependency declarations are torch>=2.7.0,<3, transformers>=5.10.1,<6, Linux vllm>=0.19.1,<0.29.0, and Windows lmdeploy>=0.17.0,<0.18. Intersect the package's Python range with the wheels actually available for these dependencies. Check the legacy platform policy before upgrading a vendor environment.

Apple Silicon defaults to llama.cpp for the VLM.

The WebUI uses Gradio 6 and bundled PDF.js preview assets; it does not require gradio-pdf. A separate UI environment can use the base package and connect with mineru-kit webui --api-url http://127.0.0.1:8000.

Upgrade an existing environment

Stop that environment's MinerU services before upgrading with the original installer. Preserve the required extras, then check the version, dependencies, and model configuration. Downloads and inference must use the same model.small_backend and model.vlm.engine; see Model Source.