Run and Validate

Most runnable PyPTO-Lib scripts use the local Golden Harness to turn a kernel definition into a correctness result. A few specialized smoke, regeneration, or external-runtime drivers call PyPTO's compile/runtime APIs directly; inspect the selected script's __main__ block and --help before assuming that it performs golden validation.

The common flow is:

PyPTO compile
    ↓
input generation
    ↓
PyTorch golden computation
    ↓
simpler runtime execution
    ↓
output validation

Use the following references:

• Compile and Runtime Workflow explains the generated artifacts and each stage in detail.
• Golden Harness introduces TensorSpec, ScalarSpec, run, run_jit, validation, and RunResult.
• Save and Replay Golden Data freezes inputs and expected outputs for repeated performance-oriented runs.

For a first end-to-end command, start with Run Your First Kernel.