Creating a Cellulose Account

Let’s start by creating an account!

Signup

Signup Form

And log back into Cellulose.

Login

Login Form

Getting to know your Home Page

All your tracked machine learning models will be displayed here in your home page.

Homepage

Uploading your first ONNX model

For this example, let’s download a ResNet-18 ONNX file and upload it as your first model.

Click on the ”+ New Model” button at the bottom right your dashboard

Create Model

then upload it in the modal form Upload Model

Uploading Model

It may take a while before the model shows up in the table view as our backend workers furiously unroll the ONNX model protobuf. We’re working on improvements to make it faster so stay tuned for more!

Visualizing your first ONNX model

Congrats! you’ve uploaded your first model onto the Cellulose platform! Let’s move on to the fun part - visualizing it :)

Click on the model name to proceed.

Navigating to ONNX Model Editor

ResNet model graph

You’ll now see the ONNX graph being rendered, where an output of a given node serve as input(s) to others in a graph runtime.

Inspecting a Node

You may click on the node itself to gather more details about the node.

Inspecting a node

A drawer should slide out from the right where you’ll also get basic information such as its name, domain or doc string, if present.

Node Drawer

Understanding Node Inputs / Outputs / Attributes

Let’s dive a little deeper into all the inputs and outputs associated with a BatchNormalization operation below as an example.

We tried to make this a little simpler by organizing them into tabs. As of this writing, these are Inputs, Outputs, Attributes, Sample Output Values and Fused. We’ll go into each of them in the sections below.

Each operator will always have a set of input(s) and output(s). The input(s) of a node may consist of:

  1. Tensor initializers with real values, and/or
  2. An input that happen to also be the output of an upstream node.

Node Inputs

Input tab

Node Outputs

Output tab

Some nodes / operators will also have attributes like in this BatchNormalization example:

Node Attributes

Viewing / Copying Node Initializer Values

Click on the “Show Values” of one of the inputs

Navigating to Node Initializers

This should bring up a modal overlay with even more details about that particular initializer.

Navigating to Node Initializers

Well done! Now you’ve dabbled with the core functionalities of the Cellulose dashboard. Feel free to browse the other documentation sections for more details.

Have questions / need help?

Please reach out to support@cellulose.ai, and we’ll get back to you as soon as possible.