Update app.py

app.py

@@ -41,6 +41,7 @@ class FinancialDataset(Dataset):
 class FinancialAnalyzer:
     def __init__(self):
         print("Initializing Analyzer...")
+        self.last_metrics = {} 
         self.initialize_models()
         print("Initialization complete!")
 
@@ -323,33 +324,92 @@ Provide:
     def generate_analysis(self, prompt):
         """Generate analysis using TinyLlama"""
         try:
-            inputs = self.llama_tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
-            
+            # Format the prompt in TinyLlama's expected format
+            formatted_prompt = f"<human>: {prompt}\n<assistant>: Let me analyze these financial metrics in detail."
+        
+            inputs = self.llama_tokenizer(
+                formatted_prompt,
+                return_tensors="pt",
+            truncation=True,
+            max_length=2048,
+            padding=True
+        )
+        
+        # Generate with adjusted parameters
             outputs = self.llama_model.generate(
                 inputs["input_ids"],
-                max_new_tokens=1024,
-                temperature=0.7,
-                top_p=0.9,
-                do_sample=True,
-                repetition_penalty=1.2,
-                no_repeat_ngram_size=3,
-                num_return_sequences=1,
-                pad_token_id=self.llama_tokenizer.eos_token_id
-            )
-            
-            analysis = self.llama_tokenizer.decode(outputs[0], skip_special_tokens=True)
-            
-            # Extract response after instruction
-            if "[/INST]" in analysis:
-                analysis = analysis.split("[/INST]")[-1].strip()
-            
-            # Validate the analysis
+            max_new_tokens=1024,
+            min_new_tokens=200,  # Ensure minimum length
+            temperature=0.8,      # Slightly increased creativity
+            top_p=0.92,          # Slightly increased diversity
+            do_sample=True,
+            repetition_penalty=1.2,
+            length_penalty=1.5,   # Encourage longer generations
+            num_return_sequences=1,
+            pad_token_id=self.llama_tokenizer.eos_token_id,
+            eos_token_id=self.llama_tokenizer.eos_token_id,
+            early_stopping=True
+        )
+        
+        # Decode and clean up the response
+            analysis = self.llama_tokenizer.decode(outputs[0], skip_special_tokens=False)
+        
+        # Extract only the assistant's response
+            if "<assistant>:" in analysis:
+                analysis = analysis.split("<assistant>:")[-1].strip()
+        
+        # Clean up any remaining tags
+            analysis = analysis.replace("<human>:", "").replace("<assistant>:", "").strip()
+        
+        # Validate output length and content
             if len(analysis.split()) < 100:
-                return "Error: Generated analysis is too short. Please try again."
+            # Fallback analysis if model generation is too short
+                analysis = self.generate_fallback_analysis(self.last_metrics)
             
             return analysis
+
+        except Exception as e:
+            print(f"Detailed error in generate_analysis: {str(e)}")
+            return self.generate_fallback_analysis(self.last_metrics)
+
+
+    def generate_fallback_analysis(self, metrics):
+        """Generate a basic analysis when the model fails"""
+        try:
+            revenue_growth = metrics['Ratios'].get('Revenue_Growth', 0)
+            net_margin = metrics['Ratios'].get('Net_Margin', 0)
+            current_ratio = metrics['Ratios'].get('Current_Ratio', 0)
+            debt_to_equity = metrics['Ratios'].get('Debt_to_Equity', 0)
+        
+            analysis = f"""
+Financial Analysis Summary:
+
+1. Revenue and Growth:
+The company shows a revenue growth of {revenue_growth:.1f}%, indicating {
+'strong' if revenue_growth > 5 else 'moderate' if revenue_growth > 0 else 'weak'} growth performance.
+
+2. Profitability:
+With a net margin of {net_margin:.1f}%, the company demonstrates {
+'strong' if net_margin > 10 else 'moderate' if net_margin > 5 else 'concerning'} profitability levels.
+
+3. Liquidity Position:
+The current ratio of {current_ratio:.2f}x suggests {
+'very strong' if current_ratio > 2 else 'adequate' if current_ratio > 1 else 'concerning'} liquidity position.
+
+4. Financial Leverage:
+With a debt-to-equity ratio of {debt_to_equity:.2f}, the company maintains {
+'conservative' if debt_to_equity < 0.5 else 'moderate' if debt_to_equity < 1 else 'aggressive'} leverage.
+
+Key Recommendations:
+1. {'Consider debt reduction' if debt_to_equity > 0.5 else 'Maintain current debt levels'}
+2. {'Focus on improving profit margins' if net_margin < 5 else 'Maintain profit efficiency'}
+3. {'Implement growth strategies' if revenue_growth < 2 else 'Sustain growth momentum'}
+
+This analysis is based on key financial metrics and standard industry benchmarks.
+"""
+            return analysis
         except Exception as e:
-            return f"Error generating analysis: {str(e)}"
+            return f"Error generating fallback analysis: {str(e)}"    
 
     def fine_tune_models(self, train_texts, train_labels, epochs=3):
         """Fine-tune the models with custom data"""
@@ -407,6 +467,7 @@ Provide:
             income_data = self.parse_financial_data(income_stmt)
             balance_data = self.parse_financial_data(balance_sheet)
             metrics = self.extract_metrics(income_data, balance_data)
+            self.last_metrics = metrics 
 
             # Get sentiment analysis
             sentiment_dict = self.get_sentiment_analysis(metrics)