Add application file

app.py

@@ -213,10 +213,10 @@ def QA_model():
         task="text-generation",
         model_kwargs={
                     "torch_dtype" : torch.float32,
-                    "low_cpu_mem_usage" : False,
+                    "low_cpu_mem_usage" :True,
                     "temperature": 0.2,
                     "max_length": 2048,
-                    #"device_map": "auto",
+                    "device_map": "auto",
                     "repetition_penalty":1.1}
             )
     print(model.model_id)