Update README.md

README.md

@@ -15,7 +15,7 @@ license: mit
   <img src="intro.png" width="100%" ></img>
   <br>
   <em>
-      1.7x Speedup and 0.5x memory consumption on ImageNet-256 generation. Top: original VAR-d30; Bottom: CoDe N=8. Speed ​​measurement does not include vae decoder
+      We partition the multi-scale inference process into a seamless collaboration between a large model and a small model.
   </em>
 </div>
 <be>