A streamlined multi-GPU server implementation.
pip install --upgrade pip
pip install uv
uv pip install -r requirements.txtpython src/server.pypython example/client.pyNow, pdf files under folder ./example/pdfs will be processed in parallel. Assuming you have 2 gpus, if you change the workers_per_device to 2, 4 pdf files will be processed at the same time!
docker build -t mineru-api:v2.0 .
Builds the Docker image for MinerU API.
docker run -d --gpus all \
-p 24008:24008 \
-v mineru-api:/app/output \
mineru-api:v2.0
Runs the container, exposing port 24008 and mounting a named volume for output.
This endpoint is used to submit PDF files for processing.
POST /predict
Request Body Example:
{
"file": "base64_encoded_pdf_content",
"options": {
"backend": "pipeline",
"lang": "en",
"method": "auto",
"formula_enable": true,
"table_enable": true
},
"file_key": "optional_unique_file_identifier"
}This endpoint allows retrieval of processed files.
GET /download/{file_key}/all.zip
GET /download/{file_key}/file.md
Replace {file_key} with the identifier returned by the /predict endpoint or a custom file_key provided in the request.
You can visit /download/ route to retrieve the output files. You might want to specify file_key in the client otherwise a random UUID would be returned by the server.
# Download all output as a zip file
curl -O http://127.0.0.1:24008/download/my_document_key/all.zip
# Retrieve only the text content as a markdown file
curl -O http://127.0.0.1:24008/download/my_document_key/file.mdGenerated files are currently hard coded to be removed in 7 days.