Skip to content

Frequently Asked Questions

If your question is not listed, try using DeepWiki's AI assistant for common issues.

For unresolved problems, join our Discord or WeChat community for support.

Why is Windows inference slow, or engine installation failing?

Windows CUDA acceleration

The base package can use ONNX / CPU and llama.cpp. For Torch small models and LMDeploy, install "mineru[full]>=4.0,<5" in the owning environment, then check Torch CUDA support, GPU drivers, and the Python range of engine wheels.

4.0 uses lmdeploy>=0.17.0,<0.18 and Transformers 5. Do not reuse old LMDeploy 0.11.x wheel recipes or bypass dependency checks. See extension modules and tiers.

What changed in commands, APIs, and configuration?

The 4.0 WebUI uses mineru-kit webui / mineru-webui, parsing APIs use /v1/*, and configuration uses config.yaml. Old /file_parse, /tasks, and model.stack cannot be reused directly. See migration.

Encountered the error ImportError: libGL.so.1: cannot open shared object file: No such file or directory in Ubuntu 22.04 on WSL2

Encountered the error ImportError: libGL.so.1: cannot open shared object file: No such file or directory in Ubuntu 22.04 on WSL2

The libgl library is missing in Ubuntu 22.04 on WSL2. You can install the libgl library with the following command to resolve the issue:

sudo apt-get install libgl1-mesa-glx

Reference: #388