Update app/llm.py

app/llm.py

@@ -114,9 +114,9 @@ async def generate(gen:GenModel):#, user: schemas.BaseUser = fastapi.Depends(cur
             seed= gen.seed,
             #chat_format="llama-2",
             stream=True,
-            #echo=True
+            echo = not True
         )
-        """
+        
         for chunk in output:
             delta = chunk['choices'][0]['delta']
             if 'role' in delta:
@@ -124,10 +124,10 @@ async def generate(gen:GenModel):#, user: schemas.BaseUser = fastapi.Depends(cur
             elif 'content' in delta:
                 print(delta['content'], end='')
             #print(chunk)
-        """
+        
         et = time()
         output["time"] = et - st
-        return output
+        print(output)
     except Exception as e:
         logger.error(f"Error in /generate endpoint: {e}")
         return JSONResponse(