updated appliation

app.py

@@ -202,12 +202,13 @@ def process_input(selected_sae, feature_number, weight_type, use_token_centroid,
     
     return result
 
+
 def gradio_interface():
     with gr.Blocks() as demo:
         gr.Markdown("# Gemma-2B SAE Feature Explorer")
         
         with gr.Row():
-            with gr.Column():
+            with gr.Column(scale=2):
                 selected_sae = gr.Dropdown(choices=["Gemma-2B layer 0", "Gemma-2B layer 6", "Gemma-2B layer 10", "Gemma-2B layer 12"], label="Select SAE")
                 feature_number = gr.Number(label="Select feature number", minimum=0, maximum=16383, value=0)
                 
@@ -225,22 +226,22 @@ def gradio_interface():
                 use_pca = gr.Checkbox(label="Introduce first PCA component")
                 pca_weight = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, label="PCA weight")
             
-            with gr.Column():
+            with gr.Column(scale=3):
+                generate_btn = gr.Button("Generate Output")
+                
                 cosine_output = gr.Column(visible=True)
                 with cosine_output:
                     gr.Markdown("100 tokens whose embeddings produce the smallest values of the ratio")
                     gr.Markdown("(cos distance from feature vector)^m/(cos distance from token centroid)^n")
-                    output_100 = gr.Textbox(label="Top 100 tokens")
+                    output_100 = gr.Textbox(label="Top 100 tokens", lines=10)
                     show_500_btn = gr.Button("Show top 500 tokens and values")
-                    output_500 = gr.Textbox(label="Top 500 tokens and values", visible=False)
+                    output_500 = gr.Textbox(label="Top 500 tokens and values", visible=False, lines=25)
                 
                 tree_output = gr.Column(visible=False)
                 with tree_output:
                     output = gr.Image(label="Tree Diagram Output")
                     neuronpedia_embed = gr.HTML(label="Neuronpedia Embed")
                     trim_slider = gr.Slider(minimum=0.00001, maximum=0.1, value=0.00001, label="Trim cutoff for cumulative probability")
-        
-        generate_btn = gr.Button("Generate Output")
 
         def update_output(selected_sae, feature_number, weight_type, use_token_centroid, scaling_factor, use_pca, pca_weight, num_exp, denom_exp, mode):
             if mode == "cosine distance token lists":
@@ -268,8 +269,4 @@ def gradio_interface():
         generate_btn.click(update_output, inputs=inputs, outputs=[output_100, output_500, cosine_output, tree_output])
         show_500_btn.click(show_top_500, outputs=output_500)
     
-    return demo
-
-if __name__ == "__main__":
-    iface = gradio_interface()
-    iface.launch()
+    return demo