painebenjamin
commented
11 months ago Hello, thank you for the suggestion!
You have impeccable timing; I just got the container registry up and running yesterday, and the docker version just passed all my automated tests.
I'm going to do a little more testing before I make an announcement post and do a full tag of 0.2.1, but it's available now if you'd like to try it. I'll be looking into what (if anything) I need to do for docker-compose support, but these are the in-progress instructions for using the container:
Running Containerized
Since this requires a GPU for effective operation, your host machine must have a GPU and be able to host GPU-accelerated Docker containers. At the current moment, this is only available using the Nvidia Container Toolkit, available for Linux machines. You must install this toolkit and then restart your the Docker daemon before you can launch Enfugue with GPU acceleration.
The containerized version includes TensorRT support.
Pulling the Container
The container is available directly off the Github Container Repository, it can be pulled like this:
docker pull ghcr.io/painebenjamin/app.enfugue.ai:latestTesting Capabilities
To check if the container is working and can communicate with your GPU, you can run the version command in the container.
docker run --rm --gpus all --runtime nvidia ghcr.io/painebenjamin/app.enfugue.ai:latest version This is the expected result:
Enfugue v.0.2.1
Torch v.1.13.1+cu117
AI/ML Capabilities:
---------------------
Device type: cuda
CUDA: Ready
TensorRT: Ready
DirectML: Unavailable
MPS: UnavailableRunning
The basic run command is:
docker run --rm --gpus all --runtime nvidia -v ${{ YOUR CACHE DIRECTORY }}:/root/.cache -p 45554:45554 ghcr.io/painebenjamin/app.enfugue.ai:latest runImport arguments are:
- Passing
--gpusis essential to tell Docker which GPUs to permit the container to user. - Pass
--runtime nvidiato force Docker to use the Nvidia runtime. - Pass
-vto mount a volume of your choice to/root/.cache, which is the parent directory where Enfugue looks for files, downloads checkpoints, etc. The directories can also be changed in the UI as needed to anything else. - Pass
-pto bind the local port45554to the container port45554.
You'll then be able to access to UI at https://app.enfugue.ai:45554. See below for information on changing the port.
Networking and Configuration
Enfugue uses a domain-based approach to allow for SSL encryption, enabling some browser features. For this reason, Enfugue uses the domain app.enfugue.ai, which resolves to the loopback address, 127.0.0.1. If you want to run Enfugue somewhere else than your local machine, you have two options:
- The easiest way to make things work is to put an entry in your
hostsfile overridingapp.enfugue.aito the IP address of the machine running Enfugue. No other configuration is necessary if you do it this way. - Configure
Enfugueto use a different domain (or no domain at all).
See Configuration for Advanced Users on how to use configuration files. You will need to ensure any configuration file passed can be read by the Docker container - this guide uses environment variables for ease-of-use.
Use a combination of SERVER_SECURE, SERVER_DOMAIN and SERVER_PORT to control how Enfugue assembles URLs and listens to requests.
- Set
SERVER_DOMAINto your desired domain name or IP address. - Set
SERVER_PORTto the desired port to listen to, or default to 45554. - Set
SERVER_SECUREto 1/True/Yes to use HTTPS, or anything else for HTTP. The default configuration uses SSL, so this must be explicitly disabled.- If you want to use HTTPS and aren't using
app.enfugue.ai, you'll need to provide your own certificate withSERVER_CERTand key withSERVER_KEY. You can also provide your own chain withSERVER.CHAIN.
- If you want to use HTTPS and aren't using
sammcj
JohanAR
Just wondering if there’s any plans to have this bundled up into a container image and perhaps docker-compose?
This would make it really easy for people to both try it out but also run it on their home servers etc…