Get Started

PyPTO-Lib holds runnable PyPTO kernels and end-to-end LLM models, plus the Golden Harness that compiles them, executes them with simpler, and compares their outputs with PyTorch references.

Follow these pages in order:

1. Install the development environment. PyPTO owns the compatible runtime, PTOAS, and PTO ISA revisions, so begin with a selected PyPTO checkout.
2. Run the first kernel on a simulator or an available NPU, and pick the platform for later examples and model kernels.
3. Read the PyPTO Coding chapter before modifying a kernel.

After the first successful run, continue with Run and Validate to understand the harness and its saved-data replay workflow.

What belongs in this repository

• examples/ contains focused programs for learning and reference.
• models/ contains model-family kernels and runnable validation entry points.
• golden/ contains the compile, runtime, and validation harness.
• docs/ contains the public workflows and technical guidance.
• build_output/ contains generated artifacts from local runs and is not source code.

Files ending in _draft.py are works in progress and are excluded from the normal runnable set.