Update README.md

README.md

@@ -90,7 +90,10 @@ Figure  6: Coefficient of Variation (standard deviation divided by mean) for Hig
 
   **Evaluation Results**
 
+**At 1.4 Billion Model Size Trained on 350 Billion Tokens**
+
 <img src="fig7.jpg" alt="fig7.jpg" style="width:1400px;"/>
+Figure 7 : Average scores of 1.4 Billion parameter models trained on 350 Billion tokens randomly sampled from state-of-the-art open datasets. Scores are averaged over 3 random seeds used for data sampling and are reported along with standard deviations.  GneissWeb performs the best among the class of large datasets.
 
 **Combining GneissWeb Components into a Winning Recipe**