add settings

app.py

@@ -141,17 +141,17 @@ st.markdown("""
 <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" integrity="sha384-Gn5384xqQ1aoWXA+058RXPxPg6fy4IWvTNh0E263XmFcJlSAwiGgFAW/dAiS6JXm" crossorigin="anonymous">
 """, unsafe_allow_html=True)
 
-strict_mode = st.radio(
-    "Query mode? Strict means all words must match in source snippet. Lenient means only some words must match.",
-    ('strict', 'lenient'))
+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'))
+    top_hits_limit = st.slider('Top hits? How many documents to use for reranking. Larger is slower but higher quality', 10, 300, 100 if torch.cuda.is_available() else 25)
+    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 run_query(query):
-    if device == 'cpu':
-        limit = 50
-        context_limit = 10
-    else:
-        limit = 100
-        context_limit = 25
+    limit = top_hits_limit or 100
+    context_limit = context_lim or 50
+
     contexts, orig_docs = search(query, limit=limit, strict=strict_mode == 'strict')
     if len(contexts) == 0 or not ''.join(contexts).strip():
         return st.markdown("""