User Guide

This section covers everything you need to run PyTorch models on the Spyre device: supported operations, tensor layouts, profiling, debugging, and worked examples.

• Running Models on Spyre
  • Using torch.compile
  • Supported Operations
  • Configuration
  • Examples
  • Troubleshooting
• Supported Operations
  • Operations Table
  • Unsupported Operations
• Tensor Layouts
  • Conceptual Overview
  • PyTorch Tensor Layouts
  • Motivation for Spyre Tensor Layouts
  • Spyre Tensor Layouts
  • DMA Encoding
  • Access Patterns
  • Default Layouts and Controlling Layouts
  • Layout Compatibility
  • Generating DCIs and SuperDSCs
  • Future Extensions
• Profiling
  • What can be profiled today
  • Toolkit layers
  • Profiling topics
  • See also
• Debugging
  • Overview
  • Step 1 — Create a Minimal Reproducer
  • Step 2 — Enable Debug Environment Variables
  • Step 3 — Examine Compiler Artifacts
  • Step 4 — Bisect Frontend vs. Backend with sendnn
  • Checklist for Filing a Bug Report
  • Quick Reference
• Examples
  • Available Examples
  • Distributed Examples
  • Provenance Audit
  • Running an Example
  • Writing Your Own Example
  • See Also