Optimising for the IPU: Computational Graph Recompilation and Executable Switching in TensorFlow

This technical note provides some strategies that can help you avoid unnecessary recompilation of code running on the IPU.

Contents

• 1. Introduction
• 2. Consideration 0: Avoiding recompilation
• 3. Consideration 1: Computational graphs
• 4. Consideration 2: Batch size
• 5. Consideration 3: Weights
• 6. Consideration 4: Concatenate
• 7. Consideration 5: Constants
• 8. Consideration 6: Deep dive
• 9. Code example
• 10. Trademarks & copyright