style

index.html

@@ -64,42 +64,6 @@
 SweetSpot v2 — Energy-per-Token demo
 </h2>
 
-<p>
-This interactive demo visualizes our analytical model for predicting the
-<b>energy consumption per generated token</b> as a function of the number of
-output tokens (<i>n</i><sub>out</sub>).
-Select one of the validated models used in the paper or load the architecture
-of any <b>dense decoder-only Transformer</b> (up to approximately 20B
-parameters) from Hugging Face to explore how the different terms contribute
-to the prediction.
-</p>
-
-<p>
-The model was fitted using empirical measurements collected on an
-<b>NVIDIA H100 GPU</b> with <b>NVIDIA TensorRT-LLM</b> under the following
-experimental configuration:
-</p>
-
-<ul>
-    <li><code>max_batch_size = 1024</code></li>
-    <li><code>max_num_tokens = 8192</code></li>
-    <li><code>num_requests = 1000</code></li>
-</ul>
-
-<p>
-Future releases will extend the analysis to additional hardware platforms
-and inference frameworks.
-</p>
-
-<p>
-For the complete methodology, experimental validation, and citation
-information, please see the paper:
-<a href="YOUR_PAPER_URL" target="_blank">
-SweetSpot v2: An Architecture-Aware Analytical Model for Predicting Energy
-Efficiency of LLM Inference
-</a>.
-</p>
-
 <h4>Validated models:</h4>
 
 <select id="preset"></select>
@@ -196,6 +160,43 @@ placeholder="tiiuae/falcon-rw-7b"
     }
 </script>
 
+
+<p>
+This interactive demo visualizes our analytical model for predicting the
+<b>energy consumption per generated token</b> as a function of the number of
+output tokens (<i>n</i><sub>out</sub>).
+Select one of the validated models used in the paper or load the architecture
+of any <b>dense decoder-only Transformer</b> (up to approximately 20B
+parameters) from Hugging Face to explore how the different terms contribute
+to the prediction.
+</p>
+
+<p>
+The model was fitted using empirical measurements collected on an
+<b>NVIDIA H100 GPU</b> with <b>NVIDIA TensorRT-LLM</b> under the following
+experimental configuration:
+</p>
+
+<ul>
+    <li><code>max_batch_size = 1024</code></li>
+    <li><code>max_num_tokens = 8192</code></li>
+    <li><code>num_requests = 1000</code></li>
+</ul>
+
+<p>
+Future releases will extend the analysis to additional hardware platforms
+and inference frameworks.
+</p>
+
+<p>
+For the complete methodology, experimental validation, and citation
+information, please see the paper:
+<a href="YOUR_PAPER_URL" target="_blank">
+SweetSpot v2: An Architecture-Aware Analytical Model for Predicting Energy
+Efficiency of LLM Inference
+</a>.
+</p>
+
 <script src="app.js"></script>
 
 </body>