better description

app.py

@@ -57,6 +57,12 @@ You can use this architecture to detect objects using textual queries. To use it
 It can be a single word or in the form of a sentence. The model is trained to recognize only 80 categories from the COCO Detection 2017 dataset. 
 Refer to <a href="https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/">this</a> website 
 or the original <a href="https://arxiv.org/pdf/1405.0312.pdf">COCO</a> paper to see the full list of categories.
+\n\n
+Best results are obtained using one of these sentences, which were used during training:
+<div class="row">
+  <div class="column">one column</div>
+  <div class="column">second column</div>
+</div>
 """
 demo = gr.Interface(
     query_image, 
@@ -72,6 +78,15 @@ demo = gr.Interface(
     ],
     cache_examples=False,
     allow_flagging = "never",
+    css = """
+    .row {
+      display: flex;
+    }
+
+    .column {
+      flex: 50%;
+    }
+    """
 )
 demo.launch()