Update app.py

app.py

@@ -90,11 +90,11 @@ If you are optimizeing for model performance and ignore inference cost this is t
 **Optimal datset size: {D_opt/Bn:.2f}B tokens**
 
 ## Your setting trade-off:
-Compared to the compute optimal model which corresponds to 100% in both cases.
+Compared to the compute optimal model.
 
-**Training compute overhead: {100 + 100*compute_overhead(kn, kd):.2f}%**
+**Training compute overhead: {100*compute_overhead(kn, kd):.2f}%**
 
-**Inference cost fraction: {kn*100:.2f}%** """
+**Inference cost savings: {100 - kn*100:.2f}%** """
     return text, fig
 
 with gr.Blocks() as demo: