Add PIGU_PINN model

[inaniloquentee]

PR Category

New Sample

Description

This PR introduces a new Physics-Informed Neural Network (PINN) sample, PIGU_Hybrid, for CFD flow reconstruction tasks.

Key Additions:

1. Model Implementation (pigu_hybrid.py):

   • Added PIGU_Hybrid class with a U-Net-like architecture.
   • Implemented SpatiallyModulatedHybridAttention for handling sparse sensor data and grid coordinates.
   • Included HybridEmbedding and GraphFeatureEncoder components.

2. Graph Extraction Script:

   • Included a run_model function within the script to demonstrate and verify the computation graph extraction using graph_net.torch.extractor.
   • Configured the script to handle multiple inputs (sensor values, sensor coordinates, grid coordinates, and base flow).

Verification:

• Tested locally. The computation graph is extracted successfully with the provided dummy inputs.

[CLAassistant]

All committers have signed the CLA.

[paddle-bot]

Thanks for your contribution!