Kuma tests browser-only execution for PyTorch models

Kuma is an project for turning a PyTorch model into one self-contained package. The package includes the model’s calculation graph, weights, browser execution code, and runtime metadata.

A small runtime loads that package and runs it directly in the browser through WebGPU, without Python, server inference, or a large runtime . The current demos use neural video representations because they are easy to test.

The longer-term target is operator networks and scientific , where a single portable file could be easier to share and run. The open design questions are whether packaging backend kernels inside the artifact is a good idea, whether this solves a real deployment problem, and how it compares with ONNX Runtime or similar systems.

Key points

  • Kuma tries to package a PyTorch model into one portable executable artifact.
  • The model can run in the browser with WebGPU instead of on a server.
  • The package includes the graph, weights, backend kernels, and runtime metadata.
  • The goal is deployment without Python or a heavyweight runtime.
  • The project is still seeking architectural feedback and comparisons with ONNX Runtime.
Read original