Deployments
Creating a new deployment
You can deploy a model from two entry points:
- From the Deployments page by clicking New Deployment
- From a completed experiment by clicking the Deploy button
From the Deployments page
When you click New Deployment, you’ll be prompted to provide a deployment name. If you don’t choose a name, one is generated for you.
You can deploy:
- A model from a completed Experiment
- A public model from Hugging Face
Deploying from a Hugging Face model
To deploy a Hugging Face model, select the Hugging Face Model option and enter the model ID in the format organization/model-name. You can either enter the ID manually or use the dropdown to browse available models.
Deploying from an experiment
To deploy a fine-tuned model from an experiment:
- Choose the experiment source
Select the Experiment option.
- Pick your experiment
Use the search box or scroll to find and select your desired experiment.
- Create the deployment
Click Create Deployment.
No further configuration is required. The deployment process begins automatically.
Deployment status and chat
Once deployment starts, the Status Overview panel shows the current state. While it’s initializing, the status reads Starting.
Once the deployment is live:
- The status reads Running
- You’ll see an endpoint URL you can use in downstream applications
- You can interact with the model directly in the Chat with the endpoint section
If the model was fine-tuned for classification, you can send inputs and get classification responses directly from the chat window.
Not all problem types support deployment. You can't deploy Multimodal Causal LM, Image Classification, or Object Detection models yet, because the vLLM inference engine used for deployments doesn't support them. See Image classification and Object detection in Experiments for problem-type-specific details.
- Submit and view feedback for this page
- Send feedback about H2O Cascade to cloud-feedback@h2o.ai