Update app.py

app.py

@@ -47,7 +47,7 @@ def perform_ocr(image):
         return_tensors="pt",
     )
     inputs = inputs.to("cuda")
-    generated_ids = model.generate(**inputs, max_new_tokens=max_tokens)
+    generated_ids = model.generate(**inputs, max_new_tokens=max_tokens, use_cache = True)
     generated_ids_trimmed = [
         out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
     ]