How To Use Riverbox Notebooks

Through this chapter, there will be a series of tutorials on how to use Riverbox.

This assumes you already have a deployment of riverbox that you are accessing through the web portal.

Execute the Sample Notebook

This is a basic tutorial to walk you through the steps to go from logging in to executing a riverbox notebook

Login

• From the landing page of the website, login.
• This could be an OAuth or username-password style credentials.
• Check with the admin of the web portal to login.

All Projects Page

• You should see a projects page as such, with one project called the "Personal Project"
• Select this Personal Project (click on it)

Project Page

• If you select the project, you should see a page as such, with a single "Demo Riverbox" and an empty filesystem
• Select this Riverbox (click on it)

Riverbox Page

• If you selected the Demo riverbox, you should see a riverbox page that looks like this

• Here, you can click execute on the top bar, and from the drop-down select Run All to execute all cells (boxes) of this Notebook
• To do a Jupyter Notebook style execution, you can also select "Debug Mode"

Execution Page

• The execution page will show the cells with an execution counter on the top
• On the second navbar, there should be a status indicator on the status of the execution
• The execution should start with maybe a slight delay, and boxes should turn green (successful, complete execution) and yellow (in progress)
• After the execution is completely finished, your screen should look something like this

Use an External Call in a Notebook

This is a basic tutorial to walk you through the steps to go from logging in to executing a riverbox notebook by calling an external call to a pre-made notebook

This is a whole separate project, with its own filesystem. It has a paper about "Delayed Fair Sharing" you can ask about on its filesystem (RAG).

Note that the "vLLM" riverbox runs a "base model", so answers may not be what you expect from "instruction tuned" models.

Login

• From the landing page of the website, login.
• This could be an OAuth or username-password style credentials.
• Check with the admin of the web portal to login.

All Projects Page

• You should see a projects page as such, with one project called the "Personal Project"
• Select this Personal Project (click on it)

Project Page

• If you select the project, you should see a page as such, with a single "Demo Riverbox" and an empty filesystem
• Make a new empty riverbox here, by clicking the NEW button

New Riverbox

• You should see a form that looks as such
• You should only need to enter a name here, and click save, or press ctrl + s

Once you do this, you should see an empty Riverbox page that looks like this

External Call

• Now it is time to add the box that calls into my external riverbox
• Make a "NEW NESTED" box, and search for "Dev's" in the search bar. You should see two options, a OpenAI one and a vLLM one.
  • You can pick either one, or both in parallel!

• You can now add an input box, with a "Default Value" of the prompt (s) you want to test out, a long with (optionally) an output box
  • Alternatively, you can add a new code box, and have a single line rbxm.output = "{whatever you want the prompt to be as a python string}"
• Your riverbox should look something like this:

Execute

• Now you can execute the riverbox
• You can execute all cells, or run it in "Notebook mode" for arbitrary cell order execution, like a computational Notebook
  • If you run in Notebook mode, remember to kill the execution when you are done
• Here is what the execution should look like