update

app.py

@@ -52,7 +52,7 @@ MARKETING_FEATURES = [
 
 class MarketingAnalyzer:
     """Main class for analyzing marketing content using Gemma Scope"""
-    
+
     def __init__(self):
         self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         # Store model size as instance variable
@@ -64,17 +64,17 @@ class MarketingAnalyzer:
         """Initialize Gemma model and tokenizer"""
         try:
             model_name = f"google/gemma-{self.model_size}"
-            
+
             # Initialize model and tokenizer with token from environment
             self.model = AutoModelForCausalLM.from_pretrained(
                 model_name,
                 device_map='auto'
             )
             self.tokenizer = AutoTokenizer.from_pretrained(model_name)
-            
+
             self.model.eval()
             logger.info(f"Initialized model: {model_name}")
-            
+
         except Exception as e:
             logger.error(f"Error initializing model: {str(e)}")
             raise
@@ -107,25 +107,25 @@ class MarketingAnalyzer:
             'categories': {},
             'recommendations': []
         }
-        
+
         try:
             # Get model activations
             inputs = self.tokenizer(text, return_tensors="pt").to(self.device)
             with torch.no_grad():
                 outputs = self.model(**inputs, output_hidden_states=True)
-            
+
             # Analyze each feature
             for feature_id, sae_data in self.saes.items():
                 feature = sae_data['feature']
                 layer_output = outputs.hidden_states[feature.layer]
-                
+
                 # Apply SAE
                 activations = self._apply_sae(
                     layer_output,
                     sae_data['params'],
                     feature.threshold
                 )
-                
+
                 # Skip BOS token and handle empty activations
                 activations = activations[:, 1:]  # Skip BOS token
                 if activations.numel() > 0:
@@ -134,7 +134,7 @@ class MarketingAnalyzer:
                 else:
                     mean_activation = 0.0
                     max_activation = 0.0
-                
+
                 # Record results
                 feature_result = {
                     'name': feature.name,
@@ -146,21 +146,21 @@ class MarketingAnalyzer:
                         feature
                     )
                 }
-                
+
                 results['features'][feature_id] = feature_result
-                
+
                 # Aggregate by category
                 if feature.category not in results['categories']:
                     results['categories'][feature.category] = []
                 results['categories'][feature.category].append(feature_result)
-            
+
             # Generate recommendations
             results['recommendations'] = self._generate_recommendations(results)
-            
+
         except Exception as e:
             logger.error(f"Error analyzing content: {str(e)}")
             raise
-        
+
         return results
 
     def _apply_sae(
@@ -191,18 +191,18 @@ class MarketingAnalyzer:
     def _generate_recommendations(self, results: Dict) -> List[str]:
         """Generate content recommendations based on analysis"""
         recommendations = []
-        
+
         try:
             # Get technical features
             tech_features = [
                 f for f in results['features'].values()
                 if f['category'] == 'technical'
             ]
-            
+
             # Calculate average technical score if we have features
             if tech_features:
                 tech_score = np.mean([f['activation_score'] for f in tech_features])
-                
+
                 if tech_score > 0.8:
                     recommendations.append(
                         "Consider simplifying technical language for broader audience"
@@ -213,7 +213,7 @@ class MarketingAnalyzer:
                     )
         except Exception as e:
             logger.error(f"Error generating recommendations: {str(e)}")
-        
+
         return recommendations
 
 def create_gradio_interface():
@@ -229,42 +229,42 @@ def create_gradio_interface():
             title="Marketing Content Analyzer (Error)",
             description="Failed to initialize. Please check if HF_TOKEN is properly set."
         )
-    
+
     def analyze(text):
         results = analyzer.analyze_content(text)
-        
+
         # Format results for display
         output = "Content Analysis Results\n\n"
-        
+
         # Overall category scores
         output += "Category Scores:\n"
         for category, features in results['categories'].items():
             if features:  # Check if we have features for this category
                 avg_score = np.mean([f['activation_score'] for f in features])
                 output += f"{category.title()}: {avg_score:.2f}\n"
-        
+
         # Feature details
         output += "\nFeature Details:\n"
         for feature_id, feature in results['features'].items():
             output += f"\n{feature['name']}:\n"
             output += f"Score: {feature['activation_score']:.2f}\n"
             output += f"Interpretation: {feature['interpretation']}\n"
-        
+
         # Recommendations
         if results['recommendations']:
             output += "\nRecommendations:\n"
             for rec in results['recommendations']:
                 output += f"- {rec}\n"
-        
+
         return output
-    
+
     # Create interface with custom theming
     custom_theme = gr.themes.Soft(
         primary_hue="indigo",
         secondary_hue="blue",
         neutral_hue="gray"
     )
-    
+
     interface = gr.Interface(
         fn=analyze,
         inputs=gr.Textbox(
@@ -283,7 +283,7 @@ def create_gradio_interface():
         theme=custom_theme
     )
     )
-    
+
     return interface
 
 if __name__ == "__main__":