updated app.py

app.py

@@ -73,14 +73,36 @@ class ModelWrapper:
                     do_sample=True,
                     pad_token_id=self.tokenizer.pad_token_id,
                     eos_token_id=self.tokenizer.eos_token_id,
-                    repetition_penalty=1.1
+                    repetition_penalty=1.2,  # Increased to reduce repetition
+                    no_repeat_ngram_size=3,  # Prevent repeating of 3-grams
+                    early_stopping=True,  # Stop when EOS token is generated
+                    stopping_criteria=None  # Will use default stopping criteria
                 )
             
             # Decode response
             response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
+            
+            # Clean up the response
             if response.startswith(prompt):
                 response = response[len(prompt):].strip()
             
+            # Remove common closure patterns
+            closures = [
+                "Best regards,",
+                "Sincerely,",
+                "Thanks,",
+                "Thank you,",
+                "Regards,",
+                "Assistant:",
+                "Human:",
+                "[Your Name]",
+                "[Student]"
+            ]
+            
+            for closure in closures:
+                if closure.lower() in response.lower():
+                    response = response[:response.lower().find(closure.lower())].strip()
+            
             generation_time = time.time() - start_time
             return response, generation_time
         except Exception as e: