Sign inSign up

bankja/docling-serve

By bankja

•Updated 6 months ago

docling-serve in a docker with lazy loading model to clear GPU VRAM

Image
0

529

bankja/docling-serve repository overview

Built with CUDA 12.8 support.

More information at https://github.com/bankjaneo/docling-serve⁠

docker-compose.yml example

services:
  docling-serve:
    image: bankja/docling-serve:cuda
    container_name: docling-serve
    ports:
      - "5001:5001"
    environment:
      - DOCLING_SERVE_ENABLE_UI=1
      - DOCLING_SERVE_MAX_SYNC_WAIT=600
      - DOCLING_SERVE_ENABLE_REMOTE_SERVICES=1
      - DOCLING_NUM_THREADS=7
      - DOCLING_SERVE_MAX_TASKS_BEFORE_RESTART=1
      - DOCLING_SERVE_FREE_VRAM_ON_IDLE=True
      - DOCLING_SERVE_UNLOAD_LLAMA_SWAP_BASE_URL=http://172.17.0.1:9292/v1
    restart: unless-stopped
    logging:
      driver: "json-file"
      options:
        max-file: "1"
        max-size: "10m"
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

Tag summary

Content type

Image

Digest

sha256:729c88827…

Size

5.4 GB

Last updated

6 months ago

docker pull bankja/docling-serve:cuda