new code

src/pdfchatbot.py

@@ -58,8 +58,8 @@ class PDFChatBot:
         self.pipeline = pipeline(
             "text-generation",
             model="meta-llama/Meta-Llama-3-8B-Instruct",
-            model_kwargs={"torch_dtype": torch.float16},
-            device_map="auto",
+            model_kwargs={"torch_dtype": torch.bfloat16},
+            device="cuda",
         )
         print("Model pipeline loaded")
 
@@ -69,7 +69,7 @@ class PDFChatBot:
         self.current_context = context
         print("Context Ready")
         print(self.current_context)
-
+    @spaces.GPU
     def create_organic_response(self, history, query):
         self.get_organic_context(query)
         messages = [