README: add live demo link, preview section, and Results cross-reference

Add a Live Demo entry to the header links table, a new Interactive Demo
section with a clickable preview image, and link the live demo from the
Results section (replacing the vague "headline plot at the top" reference).
Drop the unused figures_weather/main.jpg. Track demo_preview.jpg via LFS.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitattributes

@@ -36,7 +36,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 figures_weather/bsa.jpg filter=lfs diff=lfs merge=lfs -text
 figures_weather/bsa_runtime.jpg filter=lfs diff=lfs merge=lfs -text
 figures_weather/hurricane_tracking.jpg filter=lfs diff=lfs merge=lfs -text
-figures_weather/main.jpg filter=lfs diff=lfs merge=lfs -text
+figures_weather/demo_preview.jpg filter=lfs diff=lfs merge=lfs -text
 figures_weather/results_hres.jpg filter=lfs diff=lfs merge=lfs -text
 figures_weather/results_spectra_pareto.jpg filter=lfs diff=lfs merge=lfs -text
 figures_weather/mosaic_header.jpg filter=lfs diff=lfs merge=lfs -text

README.md

@@ -16,9 +16,9 @@ tags:
 
 # Mosaic — Block-Sparse Attention for Weather Forecasting
 
-|  📄 [**Paper**](https://arxiv.org/abs/2604.16429)  |  🤗 [**Hugging Face**](https://huggingface.co/maxxxzdn/mosaic)  |  💻 [**GitHub**](https://github.com/maxxxzdn/mosaic)  |
-| :---: | :---: | :---: |
-| ICML 2026 · arXiv:2604.16429 | Pretrained weights & model card | Source code & issue tracker |
+|  📄 [**Paper**](https://arxiv.org/abs/2604.16429)  |  🌐 [**Live Demo**](https://maxxxzdn-mosaic.static.hf.space/)  |  🤗 [**Hugging Face**](https://huggingface.co/maxxxzdn/mosaic)  |  💻 [**GitHub**](https://github.com/maxxxzdn/mosaic)  |
+| :---: | :---: | :---: | :---: |
+| ICML 2026 · arXiv:2604.16429 | Interactive forecasts & spectra | Pretrained weights & model card | Source code & issue tracker |
 
 > **(Sparse) Attention to the Details: Preserving Spectral Fidelity in ML-based Weather Forecasting Models** \
 > Maksim Zhdanov, Ana Lucic, Max Welling, Jan-Willem van de Meent · *ICML 2026*
@@ -27,6 +27,16 @@ tags:
 
 ![Spectral fidelity and skill–speed Pareto](figures_weather/results_spectra_pareto.jpg)
 
+## Interactive Demo
+
+<p align="center">
+  <a href="https://maxxxzdn-mosaic.static.hf.space/">
+    <img src="figures_weather/demo_preview.jpg" alt="Mosaic interactive demo" width="85%">
+  </a>
+</p>
+
+[**▶ Explore forecasts in your browser**](https://maxxxzdn-mosaic.static.hf.space/) — rotate the globe, step through a 10-day 16-member ensemble forecast, switch variables, and watch each member's kinetic-energy spectrum track ERA5 in real time.
+
 ## TL;DR
 
 Mosaic addresses two distinct failure modes of spectral degradation in ML-based weather prediction:
@@ -78,7 +88,7 @@ Grouped-Query Attention with ratio 4 (3 KV heads per stage), 2D RoPE on (longitu
 
 ## Results
 
-Headline plot is at the top of this page: individual ensemble members preserve realistic kinetic-energy spectra (left, 1.5°; centre, 0.25°), and Mosaic sits on the favourable end of the skill–speed–memory Pareto (right). All metrics computed at 240 h lead time, 720 initial conditions throughout the 2020 test year (1.5° benchmark) and a single 6 h forecast (0.25° benchmark).
+The headline figure at the top of the page summarizes the main result: individual ensemble members preserve realistic kinetic-energy spectra (left, 1.5°; centre, 0.25°), and Mosaic sits on the favourable end of the skill–speed–memory Pareto (right). You can explore the same forecasts and spectra interactively in the [live demo](https://maxxxzdn-mosaic.static.hf.space/). All metrics computed at 240 h lead time, 720 initial conditions throughout the 2020 test year (1.5° benchmark) and a single 6 h forecast (0.25° benchmark).
 
 On the 0.25° HRES benchmark, Mosaic competes with state-of-the-art 0.25° models despite operating at 1.5°: