---
language: en
tags:
  - physics
  - materials-science
  - graph-neural-network
  - equivariant
license: mit
---

# MPHGNet v4.1

Multi-Physics Hierarchical Graph Network (MPHGNet) v4.1 for grain-boundary energy, 
forces, stress and multi-fidelity materials property prediction.

## Model Description
- Equivariant physics backbone with e3nn irreps
- Multi-physics channels: local density, gradient flow, interaction energy, 
  global context, structural stability, Hilbert-space coherence, defect, diffusion, phonon
- Hebbian STDP plasticity module
- Triplet + quadruplet physics message passing

## Usage
```python
from huggingface_hub import hf_hub_download
import torch, importlib.util, sys

path = hf_hub_download(repo_id="harinag0204/MPHGNet-v41", filename="mphgnet_v41.py")
spec = importlib.util.spec_from_file_location("mphgnet_v41", path)
mph  = importlib.util.module_from_spec(spec); spec.loader.exec_module(mph)

model = mph.EquivariantPhysicsBackboneV41(num_layers=1).to(mph.device)
weights = hf_hub_download(repo_id="harinag0204/MPHGNet-v41", filename="model_weights.pt")
model.load_state_dict(torch.load(weights, map_location=mph.device))
model.eval()
```