Finalize

app.py

@@ -160,7 +160,7 @@ st.markdown('### Try Binder!')
 col1, _ = st.columns(2)
 with col1:
     selected_table_title = st.selectbox(
-        "Select an example table",
+        "Select an example table (We use WikiTQ examples for this demo. But task inputs can include free-form texts and images as well)",
         (
             "Estonia men's national volleyball team",
             "Highest mountain peaks of California",
@@ -183,7 +183,7 @@ st.dataframe(df)
 # Let user input the question
 with col1:
     selected_language = st.selectbox(
-        "Select a programming language",
+        "Select a target Binder program",
         ("Binder-SQL", "Binder-Python"),
     )
 if selected_language == 'Binder-SQL':
@@ -200,7 +200,7 @@ question = st.text_input(
     value=EXAMPLE_TABLES[selected_table_title][1],
 )
 
-button = st.button("Run! Solve this question by Binder program")
+button = st.button("Run Binder!")
 if not button:
     st.stop()
 

resources/demo_description.md

@@ -1,9 +1,10 @@
-This is an interactive demo of Binder based on GPT3 Codex.
-You can input a question (maybe requiring external knowledge/functionality), and a Binder program(an expression in a symbolic language that optionally includes API calls) will be generated and executed to derive the output answer.
+This is an interactive demo of [Binder](https://lm-code-binder.github.io/) based on GPT3 Codex.
+You can input a question（may not be answered by pure SQL/Python), 
+The demo will generate a Binder program (e.g., SQL/Python bound with API calls to GPT3 Codex),
+and then execute the Binder program (API calls by prompting Codex and SQL/Python by using SQL/Python interpreters) to drive the final answer.
 For more details, check out the project [website](https://lm-code-binder.github.io/).
 
 
 *Note:* 
-- Codex has query limits for openai keys, thus it may be slow in generation/execution when (high) concurrent requests occur. 
-- This page is just for demonstration, so we generate fewer program for speeding up, please check the [paper](https://arxiv.org/abs/2210.02875) and [code](https://github.com/HKUNLP/Binder) on majority vote and weight bias better result. 
-- We are trying to make the demo faster and more robust, please let us know if you have any feedback!
+- The demo might be slow when (high) concurrent requests occur because of the query limit of OpenAI GPT3 Codex. 
+- To speed up the demonstration, we only employed a simper version of Binder (e.g., no majority vote), please check the [paper](https://arxiv.org/abs/2210.02875) and [code](https://github.com/HKUNLP/Binder) for more detail.