Update app.py

app.py

@@ -157,7 +157,7 @@ def generate(
         lang_x=input_ids,
         attention_mask=attention_mask,
         max_new_tokens=30,
-        num_beams=1,
+        num_beams=3,
         do_sample=True,
         temperature=0.3,
         top_k=0,
@@ -202,7 +202,7 @@ with gr.Blocks() as demo:
     
     GitHub: [open_flamingo](https://github.com/mlfoundations/open_flamingo)
     
-    In this demo we showcase the in-context learning capabilities of the OpenFlamingo-4B model, a large multimodal model trained on top of RedPajama-INCITE-Instruct-3B-v1.
+    In this demo we showcase the in-context learning capabilities of the OpenFlamingo-4B model, a large multimodal model trained on top of RedPajama-INCITE-Instruct-3B-v1. Note that we add two additional demonstrations to the ones presented to improve the demo experience.
     The model is trained on an interleaved mixture of text and images and is able to generate text conditioned on sequences of images/text. To safeguard against harmful generations, we detect toxic text in the model output and reject it. However, we understand that this is not a perfect solution and we encourage you to use this demo responsibly. If you find that the model is generating harmful text, please report it using this [form](https://forms.gle/StbcPvyyW2p3Pc7z6).
     """
     )