add some UI improvements

app.py

@@ -136,8 +136,9 @@ st.title("Scientific Question Answering with Citations")
 
 st.write("""
 Ask a scientific question and get an answer drawn from [scite.ai](https://scite.ai) corpus of over 1.1bn citation statements.
-Answers are linked to source documents containing citations where users can explore further evidence from scientific literature for the answer. For example try:
-Are tanning beds safe to use? Does size of venture capital fund correlate with returns?
+Answers are linked to source documents containing citations where users can explore further evidence from scientific literature for the answer.
+
+For example try: Are tanning beds safe to use? Does size of venture capital fund correlate with returns?
 """)
 
 st.markdown("""
@@ -147,7 +148,7 @@ st.markdown("""
 with st.expander("Settings (strictness, context limit, top hits)"):
     strict_mode = st.radio(
         "Query mode? Strict means all words must match in source snippet. Lenient means only some words must match.",
-        ('strict', 'lenient'))
+        ('lenient', 'strict'))
     use_reranking = st.radio(
         "Use Reranking? Reranking will rerank the top hits using semantic similarity of document and query.",
         ('yes', 'no'))
@@ -158,23 +159,20 @@ with st.expander("Settings (strictness, context limit, top hits)"):
     context_lim = st.slider('Context limit? How many documents to use for answering from. Larger is slower but higher quality', 10, 300, 25 if torch.cuda.is_available() else 10)
 
 def paraphrase(text, max_length=128):
-
-  input_ids = queryexp_tokenizer.encode(text, return_tensors="pt", add_special_tokens=True)
-
-  generated_ids = queryexp_model.generate(input_ids=input_ids, num_return_sequences=5, num_beams=5, max_length=max_length)
-
-  preds = '\n'.join([queryexp_tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=True) for g in generated_ids])
-  return preds
+    input_ids = queryexp_tokenizer.encode(text, return_tensors="pt", add_special_tokens=True)
+    generated_ids = queryexp_model.generate(input_ids=input_ids, num_return_sequences=5, num_beams=5, max_length=max_length)
+    queries = set([queryexp_tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=True) for g in generated_ids])
+    preds = '\n * '.join(queries)
+    return preds
 
 
 def run_query(query):
     if use_query_exp == 'yes':
         query_exp = paraphrase(f"question2question: {query}")
         st.markdown(f"""
-        If you are not getting good results try one of:
-
-        {query_exp}
-        """)
+If you are not getting good results try one of:
+* {query_exp}
+""")
     limit = top_hits_limit or 100
     context_limit = context_lim or 10
     contexts, orig_docs = search(query, limit=limit, strict=strict_mode == 'strict')