add

index.html

@@ -172,15 +172,14 @@ <h1 class="title is-1">CFT vs. SFT</h1>
     <!-- End image carousel -->
 
 
-
     <section class="section hero is-light">
       <div class="container is-max-desktop">
         <div class="columns is-centered has-text-centered">
           <div class="column is-four-fifths">
-            <h1 class="title is-1">Comparison with other reasoning-specialized models</h1>
+            <h1 class="title is-1">Qwen2.5-Math-7B-CFT vs. Other Competitive Models</h1>
             <div class="content has-text-justified">
               <p>
-                we compare our best CFT-models with other competitive models with different scales. We expanded the evaluation benchmarks to cover broader STEM topics.
+                We compare our best 7B CFT-models with other competitive models with different scales. We expanded the evaluation benchmarks to cover broader STEM topics.
               </p>
             </div>
           </div>
@@ -205,6 +204,73 @@ <h1 class="title is-1">Comparison with other reasoning-specialized models</h1>
     </section>
     <!-- End image carousel -->
 
+    <section class="section hero is-light">
+      <div class="container is-max-desktop">
+        <div class="columns is-centered has-text-centered">
+          <div class="column is-four-fifths">
+            <h1 class="title is-1">Training Dynamics Comparison</h1>
+            <div class="content has-text-justified">
+              <p>
+                Training dynamics comparison of different methods on Qwen2.5-Math-7B across key mathematical reasoning benchmarks. We compare CFT with two SFT variants: SFT-V (trained on WebInstruct-verified) and SFT-G (trained on WebInstruct-GPT4o). The x-axis represents training steps, and the y-axis shows the accuracy on each benchmark. CFT demonstrates consistently better performance and faster convergence across most tasks
+              </p>
+            </div>
+          </div>
+        </div>
+      </div>
+    </section>
+
+     <!-- Image carousel -->
+    <section class="hero">
+        <div class="hero-body">
+            <div class="container is-max-desktop">
+                <div class="columns is-centered">
+                    <div class="column is-full">
+                        <div class="item">
+                            <!-- Your image here -->
+                            <img src="static/images/training_dynamics.png" alt="training_dynamics" />
+                        </div>
+                    </div>
+                </div>
+            </div>
+        </div>
+    </section>
+    <!-- End image carousel -->
+
+
+
+
+    <section class="section hero is-light">
+      <div class="container is-max-desktop">
+        <div class="columns is-centered has-text-centered">
+          <div class="column is-four-fifths">
+            <h1 class="title is-1"></h1>
+            <div class="content has-text-justified">
+              <p>
+                We compare the performance of 32B models across Mathematical Reasoning Benchmarks. : (1) Our most significant advantage lies in data efficiency. (2) The performance comparison across different benchmarks shows consistent advantages:
+              </p>
+            </div>
+          </div>
+        </div>
+      </div>
+    </section>
+
+     <!-- Image carousel -->
+    <section class="hero">
+        <div class="hero-body">
+            <div class="container is-max-desktop">
+                <div class="columns is-centered">
+                    <div class="column is-full">
+                        <div class="item">
+                            <!-- Your image here -->
+                            <img src="static/images/32b_comparison.png" alt="32b_comparison" />
+                        </div>
+                    </div>
+                </div>
+            </div>
+        </div>
+    </section>
+    <!-- End image carousel -->
+
 
     <!-- BibTeX citation -->
     <section class="section" id="BibTeX">