Skip to main content

Welcome to the Tinyverse

The Tinyverse is an ecosystem of tools for rapidly developing Machine Learning applications and deploying them on edge devices.

Check out our workshop from QHacks 2022 for a whirlwind overview of the Tinyverse and to see how Forge can develop and deploy Runes.

Meet The Tinyverse​

Tensors​

The most fundamental component in a machine learning application is the tensor, a multidimensional array of numbers.

You will often see tensors written like f32[1, 256, 256, 3]. This is short for a 1 x 256 x 256 x 3 tensor containing 32-bit floating point numbers.

What this tensor represents will depend entirely on the context it is used in.

Machine Learning Pipeline​

Machine learning applications work by taking information from the outside world, passing it through a neural net (often referred to as a "model"), and using the results to achieve something.

The raw information is often not in a form that is immediately usable for the model, so the data may go through one or more intermediate transformations. Common pre-processing transformations are normalizing all inputs to be between 0 and 1, or transforming an audio clip from the time domain (i.e. volume over time) to the frequency domain (signal strength per frequency).

Similarly, the model's results will often need to be post-processed before they can be given to the main application.

This entire system of inputs, outputs, models, and processing is called the Machine Learning Pipeline.

The Runefile​

A Runefile is a YAML file that declares the layout for a particular machine learning pipeline. This is consumed by the Rune compiler in order to generate a Rune.

Rune​

A Rune is the executable form of a machine learning pipeline.

The Rune compiler takes a Runefile and compiles it to a WebAssembly module that can be distributed to any device running the Rune runtime.

Forge​

Forge is a browser-based graphical UI for creating and deploying Runes.

Developers can use the Forge SDK to automatically retrieve deployed Runes from Forge for integration into their edge applications. The Forge SDK can also be configured to send telemetry back to Forge to give developers greater observability into their Rune's performance.

The Runtime​

The Rune runtime is a library that acts as the bridge between a Rune and the application that uses it. It manages loading and executing the WebAssembly module, plus provides model handlers for various Machine Learning frameworks.

What Now?​

Now you know a bit more about the Tinyverse it's time to build your first Rune!

Check out the Forge chapter for an introduction to building and deploying Runes with Forge. Otherwise if the low-level details are more your thing, you may want to read through the section on Rune.