Introduction
Svetoviz is a high-performance Python visualization engine designed for the deep inspection of multidimensional data. It renders NumPy arrays within a dynamic 3D environment, allowing to stack, rotate, and zoom into tensors down to a single scalar value.
Advanced Streaming Architecture
The engine utilizes a custom streaming protocol that synchronizes array data directly with the camera's position. When working with arrays that exceed physical RAM limits, Svetoviz leverages NumPy memory-mapped files to stream data directly from disk.
A Visual Debugger for Neural Networks
Svetoviz functions as a real-time visual debugger for machine learning. Rather than viewing static logs, you can see the model "flashing" during inference, interact with its structure, and step through live activations.
This interactive environment allows you to isolate specific activation data and point to any individual value - much like pointing a finger - to receive an immediate explanation and context of that parameter's role in the network's logic.
- Model & Activation Flow: Exploration of model architectures in a native 3D space.
- Remote & Cloud Deployment: Launch on a server for remote data browsing.
- Broad Format Support: Native rendering for NumPy, images, PyTorch, and Hugging Face.