Add library_name and improve model card with additional usage and links

README.md

@@ -1,6 +1,7 @@
 ---
 license: gpl-3.0
 pipeline_tag: graph-ml
+library_name: dualgnn
 tags:
 - gnn
 - sampling
@@ -12,6 +13,9 @@ tags:
 
 # dualGNN
 
+[![PyPI](https://img.shields.io/pypi/v/dualgnn)](https://pypi.org/project/dualgnn/)
+[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/natemacfadden/dualGNN/blob/main/tutorials/inference_demo.ipynb)
+
 dualGNN is an autoregressive message-passing GNN for sampling fine, regular
 triangulations (FRTs) of convex lattice polytopes. It operates on a
 generalization of the dual graph of a triangulation, with edges labeled by
@@ -26,6 +30,8 @@ threefolds at $h^{1,1} = 86$ (consistent with uniformity at
 $h^{1,1} = 128$) -- an order of magnitude beyond previous learned methods
 with a ~1000x smaller model.
 
+The model was presented in the paper [Sampling Triangulations and Calabi-Yau Threefolds with Autoregressive GNNs](https://arxiv.org/abs/2605.27770).
+
 ## Files
 
 | file | what it is |
@@ -58,14 +64,29 @@ path = hf_hub_download("natemacfadden/dualGNN", "reinforce.pt")
 net  = DualGNN.from_ckpt(path)
 ```
 
+### String Theory Application
+
+Pair the 2D sampler with the [NTFE algorithm](https://arxiv.org/abs/2309.10855) to sample fine, regular, star triangulations (FRSTs) of a reflexive 4D polytope:
+
+```python
+import numpy as np
+from cytools import Polytope
+from dualgnn.model import DualGNN
+from dualgnn.ntfe  import sample_ntfes
+
+verts = [[-1, -1, -1, -1], [-1, -1, -1,  3], [-1, -1,  3, -1], [-1,  3, -1, -1],
+         [ 1, -1, -1, -1], [ 1, -1, -1,  3], [ 1, -1,  3, -1], [ 1,  3, -1, -1]]
+poly = Polytope(np.array(verts, dtype=np.int64))                           # reflexive, h11 = 86
+net  = DualGNN.default()
+heights = sample_ntfes(poly, net, N=20, N_face_triangs=1_000, n_workers=4) # (20, npts) float64
+```
+
 ## Links
 
-- **Paper:** [Sampling Triangulations and Calabi-Yau Threefolds with
-  Autoregressive GNNs](https://arxiv.org/abs/2605.27770) (arXiv:2605.27770)
-- **Code / training scripts:**
-  [github.com/natemacfadden/dualGNN](https://github.com/natemacfadden/dualGNN)
-- **Benchmark protocol** (evaluate your own sampler against dualGNN):
-  [`eval/`](https://github.com/natemacfadden/dualGNN/tree/main/eval)
+- **Paper:** [Sampling Triangulations and Calabi-Yau Threefolds with Autoregressive GNNs](https://arxiv.org/abs/2605.27770) (arXiv:2605.27770)
+- **Code / training scripts:** [github.com/natemacfadden/dualGNN](https://github.com/natemacfadden/dualGNN)
+- **Interactive Demo:** [Open in Colab](https://colab.research.google.com/github/natemacfadden/dualGNN/blob/main/tutorials/inference_demo.ipynb)
+- **Benchmark protocol:** [`eval/`](https://github.com/natemacfadden/dualGNN/tree/main/eval)
 - **Archive:** [doi:10.5281/zenodo.20622920](https://doi.org/10.5281/zenodo.20622920)
 
 ## Citation
@@ -81,4 +102,4 @@ net  = DualGNN.from_ckpt(path)
   doi           = {10.48550/arXiv.2605.27770},
   url           = {https://arxiv.org/abs/2605.27770},
 }
-```
+```