CAT LAB · OPEN SOURCE

Software for seeing
what kernels are doing.

From tensor layouts to production profiling, these tools make GPU programs easier to inspect, debug, and accelerate.

05 tools 01 ecosystem tensors

Interactive visualizer

Linear Layout Viz

Explore Triton linear layouts as interactive tensor maps. Inspect matrix views, select and slice dimensions, recolor cells, and export publication-ready SVGs.

  • Triton
  • Linear layouts
  • Interactive

Local tensor viewer

Tensor Viz

Inspect N-dimensional NumPy arrays in a browser-based workspace. Slice, permute, color, compare, and organize tensors without leaving a local Python workflow.

  • NumPy
  • N-D tensors
  • Python + TypeScript

Kernel insight toolkit

Triton Viz

Visualize, profile, and check Triton kernels for memory-safety issues. Its interpreter-based workflow makes many debugging tasks available even without GPU access.

  • Visualization
  • Profiling
  • Sanitization

Lightweight Triton profiler

Proton

Capture Python context, user annotations, Triton launch metadata, and GPU kernel metrics with compact profiles that open as trees or Chrome traces.

  • GPU metrics
  • Tracing
  • Triton

Fast segment operators

Fasten

Accelerate heterogeneous graph neural networks with optimized segmented matrix multiplication and a simple interface designed to integrate with PyG.

  • HGNN
  • SegMM
  • GPU kernels