try to improve cache allocator

app.py

@@ -248,18 +248,35 @@ def profile_warmup(model_id: str):
 
 def build_alloc_plot():
     with gr.Group():
-        model = gr.Dropdown(
-            label="Model",
-            choices=["openai-community/gpt2", "google/gemma-2-2b"],
-            value="openai-community/gpt2",
-            allow_custom_value=True,
-        )
-        go = gr.Button("Run")
+        gr.Markdown("### 🚀 Cache Pre-allocator Performance Demo")
+        gr.Markdown("Compare model loading with and without transformers' caching allocator warmup. This demonstrates the memory efficiency improvements.")
+        
+        with gr.Row():
+            model = gr.Dropdown(
+                label="Model to Profile",
+                choices=[
+                    "openai-community/gpt2", 
+                    "google/gemma-2-2b",
+                    "microsoft/DialoGPT-small",
+                    "distilbert-base-uncased",
+                    "facebook/opt-125m"
+                ],
+                value="openai-community/gpt2",
+                allow_custom_value=True,
+                info="Select a model or enter a custom HuggingFace model ID"
+            )
+            go = gr.Button("🔥 Profile Memory", variant="primary")
+        
         plot = gr.LinePlot(
             x="t", y="MiB", color="mode", overlay_point=True,
-            title="from_pretrained() load: time vs CUDA memory_allocated()",
-            tooltip=["t", "MiB", "mode"], width=900, height=420
+            title="Memory Allocation Timeline: Warmup ON vs OFF",
+            tooltip=["t", "MiB", "mode"], 
+            width=900, height=450,
+            x_title="Time (seconds)",
+            y_title="Memory (MiB)"
         )
+        
+        gr.Markdown("**Note**: This demo requires GPU access. The warmup feature reduces peak memory usage during model loading.")
         go.click(profile_warmup, inputs=[model], outputs=plot)
 
 # ---------------------------
@@ -395,6 +412,51 @@ hr { border: 0; border-top: 1px solid var(--border-color); margin: 2rem 0; }
 /* Keep widgets full width */
 .gr-form, .gr-panel, .gr-block { max-width: none; }
 
+/* Terminal styling - match light mode */
+.gr-textbox textarea {
+  background: #f8fafc !important;
+  color: #1f2937 !important;
+  border: 1px solid var(--border-color) !important;
+  border-radius: 8px !important;
+  font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", monospace !important;
+  font-size: 0.9rem !important;
+  line-height: 1.5 !important;
+}
+
+.gr-textbox textarea:focus {
+  border-color: var(--link-text-color) !important;
+  box-shadow: 0 0 0 2px rgba(37, 99, 235, 0.1) !important;
+}
+
+/* Terminal output specifically */
+.gr-textbox textarea[readonly] {
+  background: #111827 !important;
+  color: #f9fafb !important;
+  border: 1px solid #374151 !important;
+  font-weight: 500 !important;
+}
+
+/* Terminal input */
+.gr-textbox:not(textarea[readonly]) textarea {
+  background: #ffffff !important;
+  color: #1f2937 !important;
+  border: 1px solid var(--border-color) !important;
+}
+
+/* Button styling */
+.gr-button {
+  background: var(--link-text-color) !important;
+  color: white !important;
+  border: none !important;
+  border-radius: 6px !important;
+  font-weight: 600 !important;
+  padding: 0.5rem 1rem !important;
+}
+
+.gr-button:hover {
+  background: #1d4ed8 !important;
+}
+
 """
 
 with gr.Blocks(css=CSS, fill_height=True, title="Interactive Blog — Transformers Feature Showcase") as demo: