How do I build my own custom model?

Overview  

You can create your own model in Dartmouth Chat using Workspace. Custom models are layered on top of existing models present in Dartmouth Chat. Once created, your custom model will be visible from the drop-down of available models. You can set a number of parameters, including system prompts, and related knowledge bases.

Adding a Custom Model  

To create a custom Model:

  • Navigate to the Workspace Models area (Choose “Workspace” from the left sidebar and select “Models”)
  • Click the “+” sign to add a new model
Adding a New Model

Creating a Custom Model  

From the resulting screen, enter information about your model. There are a number of configuration options:

Creating a New Model
  • Name. The name that will appear in the model drop-down.
  • Base Model (From). The underlying Dartmouth Chat model that will be used to construct responses. As each model has different strengths, you will likely need to experiment to find the best underlying model to meet your needs.
  • Description. A brief description that will appear as a tooltip next to the model’s name and in the header when your model is selected.
  • Image/Avatar. An image/icon providing context.
  • Tags. Custom tags that can be used to identify and group similar models.
  • Visibility. By default, the visibility of your model is set as Private.
  • System Prompt. These are guidelines and rules for the model’s responses. Your system prompt should address things like role, scope, response format, and rules. Consider using Dartmouth Chat to develop an initial draft of your system prompt.
  • Prompt Suggestions. Suggested prompts to get user conversations started.
  • Knowledge. If you have existing, relevant knowledge bases to provide context for your model, add them here. These user-curated collections will provide context for the model’s response.
  • Capabilities. Based on the capabilities of the underlying Dartmouth Chat model (e.g., Vision, Image Generation, etc.), choose what capabilities your model will expose to the user.

Using your Model  

Once you’ve saved your model, you can access it from the drop-down list of available models.

Responses will be identified by the image/avatar, name, and description provided; shaped by your system prompt; and informed by any knowledge bases you attached to the model.

Accessing your New Model Reviewing Results from your New Model

Sharing your Knowledge Base or Model  

If you want to share your custom knowledge base or model with others, please reach out to ai.support@dartmouth.edu. We will work with you to find an appropriate way to share your model with targeted users or groups.

Our preferred method for now is to construct a direct link to your Workspace Model that any Dartmouth Chat user can access; this can in turn be shared with a specific group of people via email or Slack. We are carefully considering our approach to the tool’s default sharing functionality, which adds shared custom Workspace Models to other staff members’ list of available models. We are currently working to define an easy way to share your creations without risking cluttering everyone else’s workspace.

Again, please reach out if you want to share your Workspace creations. Self-managed sharing is currently still a work in progress.

Didn’t find what you needed? Please reach out to research.computing@dartmouth.edu.