Update app.py

app.py

@@ -6,6 +6,7 @@ import uvicorn
 import huggingface_hub
 import re
 from dotenv import load_dotenv
+import spaces
 
 load_dotenv()
 
@@ -85,6 +86,7 @@ class ChatRequest(BaseModel):
     top_p: float = 0.95
     temperature: float = 0.7
 
+@spaces.GPU(duration=0)
 def generate_chat_response(request, model_data):
     try:
         user_input = normalize_input(request.message)
@@ -109,19 +111,20 @@ def remove_duplicates(text):
     text = text.replace('[/INST]', '')
     lines = text.split('\n')
     unique_lines = list(dict.fromkeys(lines))
-    return '\n'.join(unique_lines).strip()
+    return '\n'.join(unique_lines)
 
 def remove_repetitive_responses(responses):
     seen = set()
     unique_responses = []
     for response in responses:
-        normalized_response = normalize_input(response['response'])
+        normalized_response = remove_duplicates(response['response'])
         if normalized_response not in seen:
             seen.add(normalized_response)
             unique_responses.append(response)
     return unique_responses
 
 @app.post("/generate")
+@spaces.GPU(duration=0)
 async def generate(request: ChatRequest):
     try:
         responses = []
@@ -145,7 +148,7 @@ async def generate(request: ChatRequest):
             "all_responses": responses
         }
     except Exception:
-        raise HTTPException(status_code=500, detail="Internal Server Error")
+        pass
 
 def select_best_response(responses):
     return responses[0] if responses else {}