Improve response formatting and generation parameters

app.py

@@ -66,15 +66,21 @@ def chat_completions():
         inputs = tokenizer(formatted_prompt, return_tensors="pt")
         outputs = model.generate(
             inputs["input_ids"], 
-            max_new_tokens=200, 
+            max_new_tokens=256, 
             temperature=0.7,
             do_sample=True,
             top_p=0.9,
-            num_return_sequences=1
+            num_return_sequences=1,
+            repetition_penalty=1.15,
+            pad_token_id=tokenizer.pad_token_id,
+            eos_token_id=tokenizer.eos_token_id
         )
 
-        full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        response_text = full_response.replace(formatted_prompt, "").strip()
+        full_response = tokenizer.decode(outputs[0], skip_special_tokens=False)
+        
+        # Gemma formatındaki tag'leri temizle
+        response_text = full_response.split("<start_of_turn>model\n")[-1]
+        response_text = response_text.replace("<end_of_turn>", "").strip()
         
         # Vapi formatı
         vapi_response = {