GraphTensors are a fundamental part of LuminAIR, serving as references to nodes within the computational graph. They provide a high-level interface for defining machine learning models and operations, ensuring that all computations are checked at compile time.

​

Creating a GraphTensor

To get started, you can create a GraphTensor by initializing a graph and adding tensor nodes:

let mut cx = Graph::new(); // Initialize a new computational graph
let a: GraphTensor<R1<3>> = cx.tensor(); // Create a new tensor node in the graph

Here:

• Graph::new() initializes a new computational graph.
• GraphTensor<R1<3>> represents a tensor node with metadata about its shape (in this case, rank-1 with size 3).

​

Performing Operations with GraphTensors

Once you have created GraphTensors, you can perform various linear algebra operations, similar to libraries like PyTorch:

let b = a.exp().sqrt(); // Apply exponentiation and square root operations
let c = b + a; // Add tensors b and a

Operations like b + a do not consume the original tensors (a and b). Both remain available for reuse in subsequent operations. This is made possible because GraphTensor is a lightweight tracking structure that implements the Copy trait.

This design allows operations on GraphTensor to construct a computational graph without immediate execution, so no values are assigned to the GraphTensor at this stage. Actual computations are postponed until cx.gen_trace() is invoked, which enables the compiler to optimize the execution process effectively.

For more details on how to build, run, and prove graphs, check out this guide.