Update README.md

README.md

@@ -16,10 +16,8 @@ superior performance in understanding, generating, and interacting in Russian co
 
 
 ## Benchmarks
-Llama3-70B-EnSecAI-Ru-Chat outperforms the original Llama-3-70B-Instruct on the MERA benchmark by 2.4 (57.0 vs 54.6). 
+Llama3-70B-EnSecAI-Ru-Chat outperforms the original Llama-3-70B-Instruct on the MERA benchmark by 2.4% (57.0 vs 54.6). 
 Moreover, it outperforms all other Russian-language models present in this benchmark.
-
-Our model outperforms the original on the MERA benchmark by 2.6%.
 <table>
   <tr>
    <td><strong>Category</strong>