Getting Started with Jozu Hub
Jozu Hub is a service catalogue for AI/ML projects that reduces development time by >31% while increasing security. You can learn more about what Jozu Hub is and what users are doing with it in our overview.
In this guide you'll learn how to:
- Use the Top 10 Jozu Hub features and discover the richness of KitOps ModelKits
- Create your own ModelKit by importing from Hugging Face
- Create your own ModelKit from your computer and share it on Jozu Hub
INFO
For the best experience, sign up for free. Registered users can import from Hugging Face, see security reports, and generate deployable containers automatically.
Top 10 Jozu Hub Features
- Hundreds of secure public models: Selecting one of the ModelKits on the Discover or Browse page will bring you into the repository view
- Tags: Each ModelKit in a repository has a "Tag" that is in green text beside the repository name
TIP
A tag is an alias for a ModelKit. You can think of it as the ModelKit's label. Often tags will be used to indicate a version (e.g., latest
or 1.4
), but they can also be used to indicate specifics about what's in the ModelKit - like the quantization of a model (e.g., q8_0
or q4_0
), or the purpose of a dataset (e.g., validation
or training
).
You can change which tag you're viewing by using the drop down beside the Tag name.
- Multiple model versions: Click the down arrow to see other Tags (versions) of the model
- Quickly pull any model: Clicking the "Pull Tag" button will copy the Kit CLI command needed to download the ModelKit to your computer (learn more about the Kit CLI)
- Peek into the ModelKit without pulling it: The box to the right gives a quick view of what's inside the ModelKit (ModelKits can contain models, datasets, parameters, metrics, code, or anything else needed to share and reproduce an AI/ML project). This gives you an overview of what you'll get before you pull a large ModelKit.
- Rich metadata for the whole team: The series of tabs below the "Pull Tag" text box feature the model card, contents, deployment and security information, and the ModelKit diff view.
ModelKit Contents: See metadata on everything inside the ModelKit - including parameters and metrics. A good example is here.
Deploy: Jozu Hub automatically generates a Docker container or Kubernetes deployment YAML for any ModelKit. For Docker containers simply copy the
docker run
command we provide; for Kubernetes copy the YAML we generate. You can learn more about this in our deployment docs.
- Security Report: Every ModelKit is scanned for vulnerabilities. This process can take some time for large models so don't worry if your security report isn't complete when you first check on a new model. There's a good example here, but while everyone can see the security summary, only logged in users can see the security report details.
- ModelKit Diff: The Jozu Hub generates diffs for each ModelKit tag. If your repository has more than one tag in it you can compare the changes between them with the diff feature.
Import from Hugging Face
The fastest way to start is by importing a model you want to use from Hugging Face. Jozu Hub will automatically package it as a shareable and runnable ModelKit and scan it for vulnerabilities.
Make sure you're logged in - a free account is needed.
Click the "Add Repository" button and select "Import from Hugging Face" in the drop-down.
- Paste in the Hugging Face URL for the model you want. If the model is token-protected you'll need to add your token otherwise the rest of the fields will automatically populate.
- Click "Import"
Importing large models can take time, but we'll email you when they're ready.
Creating a ModelKit
You can create a new repository and use it to share ModelKits you've built yourself.