scheitelpunk/GASM_weights

🎯 GASM Weights

Geometric Attention for Spatial & Mathematical Understanding

Neural architecture for SE(3)-invariant spatial reasoning from natural language

---

🧠 About GASM

GASM stands for Geometric Attention for Spatial & Mathematical Understanding

This repository contains the pretrained model weights used in the live Hugging Face Space to map input text into structured 3D spatial representations.

🔬 Technical Components

• 🗣️ NLP: spaCy-based entity & relation parsing
• 📐 Geometry: Lie group operations in SE(3) space
• 🌀 Optimization: Riemannian curvature optimization

---

📂 Repository Contents

File	Description
gasm_weights.pth	PyTorch weights file used by the core GASM engine

---

🚀 Quick Start

Load the weights into your own pipeline using huggingface_hub:

from huggingface_hub import hf_hub_download
import torch

# Download and load weights
weights_path = hf_hub_download("scheitelpunk/gasm-weights", "gasm_weights.pth")
model.load_state_dict(torch.load(weights_path))

---

📄 License

Released under CC-BY-NC 4.0 - For research and non-commercial use only. All rights reserved, Versino PsiOmega GmbH

📚 Citation

If you use GASM or its weights, please cite:

@misc{gasm2025,
  title={GASM: Geometric Attention for Spatial Understanding},
  author={Michael Neuberger, Versino PsiOmega GmbH},
  year={2025},
  url={https://huggingface.co/spaces/scheitelpunk/GASM}
}

---

🔗 Related Repositories

• 🔬 GASM Space - Live demo
• 💻 Source Code - Implementation

---

Made with ❤️ for geometric deep learning

Versino PsiOmega GmbH, http://psiomega.versino.de