TorchSharpVisual: Neural network architecture visualization for TorchSharp

[JacobGoodchild]

Hey everyone!

I wanted to share a project I've been building for the .NET AI ecosystem called TorchSharpVisual.

If you've used Python packages like torchview or visualkeras, you know how valuable it is to visualize your models. TorchSharpVisual brings that functionality natively to TorchSharp.

Here's what it does.

---

Automatic Model Inspection: Recursively walks sub-modules with zero manual code annotations.

Real Dataflow Edge Tracking: Tracks tensor identities during a dummy execution pass so branching/merging topologies (e.g., Residual connections) render accurately rather than flattening into straight lines.

Shape & Parameter Insights: Annotates input/output shapes and parameter counts per layer, along with a top-level model summary header.

Exporting: Renders straight to PNG, SVG, or raw Graphviz .dot files.

---

NuGet:
https://www.nuget.org/packages/TorchSharpVisual

---

Github:
https://github.com/JacobGoodchild/TorchSharpVisual?tab=readme-ov-file

---

Or use bash:
dotnet add package TorchSharpVisual --version 0.2.0

[GeorgeS2019]

@JacobGoodchild
See if your approach could also be used to check the export of TorchSharp to Onnx
https://github.com/gluschenko/onnxify/blob/main/.docs/nuget/Onnxify.TorchSharp.md