Skip to content

docker.md

Latest commit

 

History

History
executable file
·
90 lines (65 loc) · 3.23 KB

docker.md

File metadata and controls

executable file
·
90 lines (65 loc) · 3.23 KB
Raw

DedrisGenAI on Docker

Run the full DedrisGenAI app in a container — the Python+PyTorch engine and the PHP web UI. The image is based on NVIDIA CUDA 12.4; see docker/Dockerfile, docker/entrypoint.sh and engine/requirements_versions.txt.

The container runs the engine internally on 127.0.0.1:7866 and serves the PHP web UI on port 8888 (the only exposed service) — the same UI you get from the desktop launchers, not the legacy Gradio UI. Open http://localhost:8888 once it's up.

Requirements

  • An NVIDIA GPU with proprietary drivers + the NVIDIA Container Toolkit
  • Docker (with Compose) or Podman

Quick start (Docker Compose)

From the repository root:

docker compose -f docker/docker-compose.yml up --build

Then open http://localhost:8888. The first generation downloads the selected model into the persistent volume (a few GB; one time per model).

Quick start (Docker)

Build (context is the repo root so both engine/ and webui/ are included):

docker build -f docker/Dockerfile -t dedrisgenai .

Run:

docker run --gpus all -p 8888:8888 -v dedrisgenai-data:/content/data -it dedrisgenai

Quick start (Podman)

podman build -f docker/Dockerfile -t dedrisgenai .
podman run --device=nvidia.com/gpu=all -p 8888:8888 -v dedrisgenai-data:/content/data -it dedrisgenai

Persistence

Models and generated images live in the dedrisgenai-data volume (mounted at /content/data). entrypoint.sh symlinks engine/models → /content/data/models and engine/outputs → /content/data/outputs, seeds the models folder with the small support files shipped in the image (prompt-expansion config/tokenizer), and lets the engine download weights into the volume on first use. The volume persists across container restarts.

Environment variables

Variable Default Meaning
DEDRIS_UI_PORT 8888 Port the PHP UI listens on (exposed).
DEDRIS_ENGINE_PORT 7866 Internal engine port (not exposed).
DEDRIS_HOST 0.0.0.0 Bind address for the PHP UI.
DATADIR /content/data Persistent volume location (models + outputs).

Paths inside the container

Path Details
/content/app The app (engine/ + webui/).
/content/app/engine/models.org The shipped models support files; copied into the volume-backed engine/models on boot (existing files are not overwritten).
/content/data/models Persistent models (symlinked to engine/models).
/content/data/outputs Persistent outputs (symlinked to engine/outputs).

Updating

git pull
docker compose -f docker/docker-compose.yml build --no-cache
docker compose -f docker/docker-compose.yml up

Notes

  • The model weights and the provisioned runtimes/ are excluded from the build context via .dockerignore, so the image stays lean and weights are fetched at runtime.
  • The MPS backend (Apple Silicon) is not supported in Docker — on a Mac use the start.command launcher instead.
  • docker compose -f docker/docker-compose.yml up -d runs detached; follow logs with docker compose -f docker/docker-compose.yml logs -f.