MatViT README edits

PiperOrigin-RevId: 575346989

scenic/projects/README.md

@@ -76,10 +76,10 @@
     > method to train spatially-aware vision transformer features.
 
 *   [MatViT](matvit)
-    > MatViT ([paper](https://arxiv.org/abs/2310.07707)) is a nested ViT
-    > architecture designed to offer elasticity in a variety of deployment
-    > constraints, where each Feed Forward Network (FFN) block of a MatViT
-    > model is jointly optimized with a few nested smaller FFN blocks.
+    > MatViT is a MatFormer ([paper](https://arxiv.org/abs/2310.07707)) based
+    > nested ViT architecture designed to offer elasticity in a variety of
+    > deployment constraints, where each Feed Forward Network (FFN) block of a
+    > MatViT model is jointly optimized with a few nested smaller FFN blocks.
 
 *   [MBT](mbt)
 

scenic/projects/matvit/README.md

@@ -1,10 +1,10 @@
-### Matryoshka Vision Transformers
+### Matryoshka Vision Transformers (MatViT)
 
 This project implements MatViT in Scenic.
 
-This is implemented according to the MatViT part in the MatFormer publication, for details see
+This is implemented according to the MatViT adaptation in the MatFormer paper, for details see
 Devvrit et al., "MatFormer: Nested Transformer for Elastic Inference",
-https://arxiv.org/abs/2310.07707. Further optimizations can be applied to speed up the training.
+https://arxiv.org/abs/2310.07707. This implementation provides the basic proof of concept and further optimizations can be applied to speed up the training.
 
 If you are interested in using the code and/or pretrained models, please contact
 francischen@google.com and kusupati@google.com.