Commit
Β·
b78304c
0
Parent(s):
Initial commit with Git LFS tracking
Browse files- .gitattributes +1 -0
- README.md +96 -0
.gitattributes
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-nc-4.0
|
| 3 |
+
tags:
|
| 4 |
+
- geometric deep learning
|
| 5 |
+
- SE(3)
|
| 6 |
+
- spatial reasoning
|
| 7 |
+
- AI geometry
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
<div align="center">
|
| 11 |
+
|
| 12 |
+
# π― GASM Weights
|
| 13 |
+
## Geometric Attention for Spatial & Mathematical Understanding
|
| 14 |
+
|
| 15 |
+
[](https://creativecommons.org/licenses/by-nc/4.0/)
|
| 16 |
+
[](https://huggingface.co/spaces/scheitelpunk/GASM)
|
| 17 |
+
|
| 18 |
+
*Neural architecture for SE(3)-invariant spatial reasoning from natural language*
|
| 19 |
+
|
| 20 |
+
</div>
|
| 21 |
+
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
+
## π§ About GASM
|
| 25 |
+
|
| 26 |
+
**GASM** stands for **Geometric Attention for Spatial & Mathematical Understanding**
|
| 27 |
+
|
| 28 |
+
This repository contains the pretrained model weights used in the live [Hugging Face Space](https://huggingface.co/spaces/scheitelpunk/GASM) to map input text into structured 3D spatial representations.
|
| 29 |
+
|
| 30 |
+
### π¬ Technical Components
|
| 31 |
+
|
| 32 |
+
- π£οΈ **NLP**: spaCy-based entity & relation parsing
|
| 33 |
+
- π **Geometry**: Lie group operations in **SE(3)** space
|
| 34 |
+
- π **Optimization**: Riemannian curvature optimization
|
| 35 |
+
|
| 36 |
+
---
|
| 37 |
+
|
| 38 |
+
## π Repository Contents
|
| 39 |
+
|
| 40 |
+
| File | Description |
|
| 41 |
+
|------|-------------|
|
| 42 |
+
| `gasm_weights.pth` | PyTorch weights file used by the core GASM engine |
|
| 43 |
+
|
| 44 |
+
---
|
| 45 |
+
|
| 46 |
+
## π Quick Start
|
| 47 |
+
|
| 48 |
+
Load the weights into your own pipeline using `huggingface_hub`:
|
| 49 |
+
|
| 50 |
+
```python
|
| 51 |
+
from huggingface_hub import hf_hub_download
|
| 52 |
+
import torch
|
| 53 |
+
|
| 54 |
+
# Download and load weights
|
| 55 |
+
weights_path = hf_hub_download("scheitelpunk/gasm-weights", "gasm_weights.pth")
|
| 56 |
+
model.load_state_dict(torch.load(weights_path))
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
---
|
| 60 |
+
|
| 61 |
+
## π License
|
| 62 |
+
|
| 63 |
+
Released under **CC-BY-NC 4.0** - For research and non-commercial use only. All rights reserved, Versino PsiOmega GmbH
|
| 64 |
+
|
| 65 |
+
### π Citation
|
| 66 |
+
|
| 67 |
+
If you use GASM or its weights, please cite:
|
| 68 |
+
|
| 69 |
+
```bibtex
|
| 70 |
+
@misc{gasm2025,
|
| 71 |
+
title={GASM: Geometric Attention for Spatial Understanding},
|
| 72 |
+
author={Michael Neuberger, Versino PsiOmega GmbH},
|
| 73 |
+
year={2025},
|
| 74 |
+
url={https://huggingface.co/spaces/scheitelpunk/GASM}
|
| 75 |
+
}
|
| 76 |
+
```
|
| 77 |
+
|
| 78 |
+
---
|
| 79 |
+
|
| 80 |
+
## π Related Repositories
|
| 81 |
+
|
| 82 |
+
- π¬ **[GASM Space](https://huggingface.co/spaces/scheitelpunk/GASM)** - Live demo
|
| 83 |
+
- π» **[Source Code](https://github.com/scheitelpunk/GASM-Huggingface)** - Implementation
|
| 84 |
+
|
| 85 |
+
---
|
| 86 |
+
|
| 87 |
+
## π Contribute
|
| 88 |
+
|
| 89 |
+
This is active research! If you'd like to extend GASM (e.g., add spatial relationships, languages, domains), feel free to fork and build upon it.
|
| 90 |
+
|
| 91 |
+
<div align="center">
|
| 92 |
+
|
| 93 |
+
**Made with β€οΈ for geometric deep learning**
|
| 94 |
+
|
| 95 |
+
</div>
|
| 96 |
+
|