docs(README): add attention demo GIF + data-efficiency headline

Add assets/haaland_full_attn_blk20.gif (LFS) self-attention visualization; track *.gif via LFS; add topline noting ~1/3 of DINOv3's data (0.5B vs 1.7B) with competitive performance.

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text

README.md

@@ -18,6 +18,14 @@ backbone. A single **3D-convolutional tokenizer** ingests both modalities — an
 1-frame clip (`T=1`), a video is `T>1` — so the same weights produce **dense patch-level
 features** and a **language-aligned global (CLS) representation**.
 
+Trained on **~1/3 the data of DINOv3** (0.5B vs 1.7B samples), it still reaches competitive
+performance across image and video benchmarks — and leads on DAVIS video tracking.
+
+<p align="center">
+  <img src="assets/haaland_full_attn_blk20.gif" width="480" alt="Self-attention visualization on a video clip"/>
+</p>
+<p align="center"><em>Self-attention over patch tokens (block 20) on a video clip — the encoder attends coherently to the subject across frames.</em></p>
+
 - **Architecture**: ViT-7B (embed 4096 / depth 40 / heads 32), patch 16, 3D axial RoPE
   (`base=100`), SwiGLU FFN, LayerScale, per-head QK-norm, gated attention, 4 register tokens.
 - **Tokenizer**: `Conv3d(kernel=stride=(tubelet, patch, patch))` — image `(B,3,H,W)` → `T=1`,