Update app.py

app.py

@@ -69,15 +69,34 @@ class StopOnTokens(StoppingCriteria):
 
 
 def predict(message, history):
-    text = vector_search(message)
-    message = message + "\n\n### Source ###\n" + text
 
-    print(message)
+    #For now, we only query the vector database once, at the start.
+    if len(history) == 0:
+        source_text = vector_search(message)
+    
     history_transformer_format = history + [[message, ""]]
+
+    print(history_transformer_format)
     stop = StopOnTokens()
 
-    messages = "".join(["".join(["<|start_header_id|>user<|end_header_id|>\n\n"+item[0], "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"+item[1]])
-                for item in history_transformer_format])
+    messages = []
+    id_message = 1
+    total_message = len(history_transformer_format)
+    for item in history_transformer_format:
+
+        #Once we target the ongoing post we add the source.
+        if id_message == total_message:
+            question = "<|start_header_id|>user<|end_header_id|>\n\n"+ item[0] + "\n\n### Source ###\n" + source_text
+        else:
+            question = "<|start_header_id|>user<|end_header_id|>\n\n"+ item[0]
+        answer = "<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"+item[1]
+        result = "".join([question, answer])
+        messages.append(result)
+        id_message = id_message + 1
+
+    messages = "".join(messages)
+
+    print(messages)
 
     messages = system_prompt + messages