L04 Building and Validating IR

Metadata

• Lecture ID: L04
• Track: B (IR and Execution)
• Tier: 1-2
• Duration: 25 minutes
• Prerequisites: L01-L03

Learning Objectives

1. Move from pipeline graph to validated IR.
2. Read IR node/edge/topology fields and explain their purpose.
3. Explain how execution build policies reduce executor count.

Core Concept

• Mental model: validation converts authored graph into explicit runtime contract.
• Key pitfall: treating IR as debug noise instead of the execution source of truth.

Live Demo Mapping

• Primary runnable files:
• examples/tutorial/b_ir_and_execution/01_context_graph.py
• examples/tutorial/b_ir_and_execution/02_ir_validation.py
• Optional extension: examples/tutorial/b_ir_and_execution/03_execution_build.py

Runnable Commands

Run from repository root:

pixi run python -m examples.tutorial.b_ir_and_execution.01_context_graph
pixi run python -m examples.tutorial.b_ir_and_execution.02_ir_validation
pixi run python -m examples.tutorial.b_ir_and_execution.03_execution_build

What To Observe

• Graph output shows node list, edge list, sources, sinks, and cycle groups.
• Validation prints Validation successful and JSON IR with nodes, edges, topology.
• Execution build shows partition count and node reduction summary.

Failure Drill

• Introduce a mismatched port map or type mismatch and rerun validation.
• Confirm failure happens before runtime and identify which contract check caught it.

Exercise

• Required task: capture one IR JSON and annotate 3 fields that drive runtime behavior.
• Stretch task: compare conservative vs aggressive policy reduction and explain tradeoff.

Evaluation Rubric

• Pass criteria: learner can explain why compile/validation stage prevents runtime ambiguity.
• Common failure signature: learner cannot map an IR edge back to source/destination flow ports.

Follow-up

• Next lecture: l05_runtime_backends_and_execution_lifecycle.md
• Related tutorials: b_ir_and_execution/04_rt_execution.py