Trim README

README.md

@@ -60,8 +60,8 @@ configs:
 </p>
 
 </div>
-The frozen **evaluation split** of VBVR-Pro. 100 parameterized reasoning tasks, 5 held-out instances each, in both a video (I2V) and an interleaved-image setting.
 
+The frozen **evaluation split** of VBVR-Pro. 100 parameterized reasoning tasks, 5 held-out instances each, in both a video (I2V) and an interleaved-image setting.
 
 ## At a glance
 
@@ -133,15 +133,6 @@ cd VBVR-Pro-Bench && tar xzf VBVR-Pro-Bench-Video.tar.gz
 
 Generate one video (or image sequence) per instance from `first_frame.png` + `prompt.txt`, then score with [VBVR-Pro-Bench](https://github.com/Video-Reason/VBVR-Pro-Bench).
 
-For the full project — training data, recipes and models — see [VBVR-Pro](https://github.com/Video-Reason/VBVR-Pro).
-
-> This benchmark is **1024 × 1024**, as generated. The training splits
-> ([SFT-Video](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Video),
-> [SFT-Image](https://huggingface.co/datasets/Video-Reason/VBVR-Pro-SFT-Image))
-> are downscaled to 512 × 512, so training and evaluation resolutions differ by
-> design. Evaluate at 1024 × 1024 regardless of the resolution a model was
-> trained at, or state the deviation.
-
 ## License
 
 VBVR-Pro source code, scripts, configuration files and task-specific scoring