refine agent

agent/component/categorize.py

@@ -73,7 +73,7 @@ class Categorize(Generate, ABC):
 
     def _run(self, history, **kwargs):
         input = self.get_input()
-        input = "Question: " + ("; ".join(input["content"]) if "content" in input else "") + "Category: "
+        input = "Question: " + (list(input["content"])[-1] if "content" in input else "") + "\tCategory: "
         chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
         ans = chat_mdl.chat(self._param.get_prompt(), [{"role": "user", "content": input}],
                             self._param.gen_conf())

agent/component/generate.py

@@ -101,7 +101,7 @@ def _run(self, history, **kwargs):
         prompt = self._param.prompt
 
         retrieval_res = self.get_input()
-        input = ("  - " + "\n  - ".join(retrieval_res["content"])) if "content" in retrieval_res else ""
+        input = ("  - "+"\n  - ".join([c for c in retrieval_res["content"] if isinstance(c, str)])) if "content" in retrieval_res else ""
         for para in self._param.parameters:
             cpn = self._canvas.get_component(para["component_id"])["obj"]
             _, out = cpn.output(allow_partial=False)

agent/component/rewrite.py

@@ -33,7 +33,7 @@ def __init__(self):
     def check(self):
         super().check()
 
-    def get_prompt(self):
+    def get_prompt(self, conv):
         self.prompt = """
         You are an expert at query expansion to generate a paraphrasing of a question.
         I can't retrieval relevant information from the knowledge base by using user's question directly.     
@@ -43,6 +43,40 @@ def get_prompt(self):
         And return 5 versions of question and one is from translation.
         Just list the question. No other words are needed.
         """
+        return f"""
+Role: A helpful assistant
+Task: Generate a full user question that would follow the conversation.
+Requirements & Restrictions:
+  - Text generated MUST be in the same language of the original user's question.
+  - If the user's latest question is completely, don't do anything, just return the original question.
+  - DON'T generate anything except a refined question.
+
+######################
+-Examples-
+######################
+# Example 1
+## Conversation
+USER: What is the name of Donald Trump's father?
+ASSISTANT:  Fred Trump.
+USER: And his mother?
+###############
+Output: What's the name of Donald Trump's mother?
+------------
+# Example 2
+## Conversation
+USER: What is the name of Donald Trump's father?
+ASSISTANT:  Fred Trump.
+USER: And his mother?
+ASSISTANT:  Mary Trump.
+User: What's her full name?
+###############
+Output: What's the full name of Donald Trump's mother Mary Trump?
+######################
+# Real Data
+## Conversation
+{conv}
+###############
+    """
         return self.prompt
 
 
@@ -56,14 +90,12 @@ def _run(self, history, **kwargs):
             self._loop = 0
             raise Exception("Sorry! Nothing relevant found.")
         self._loop += 1
-        q = "Question: "
-        for r, c in self._canvas.history[::-1]:
-            if r == "user":
-                q += c
-                break
+
+        conv = self._canvas.get_history(4)
+        conv = "\n".join(conv)
 
         chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
-        ans = chat_mdl.chat(self._param.get_prompt(), [{"role": "user", "content": q}],
+        ans = chat_mdl.chat(self._param.get_prompt(conv), [{"role": "user", "content": "Output: "}],
                             self._param.gen_conf())
         self._canvas.history.pop()
         self._canvas.history.append(("user", ans))

api/apps/canvas_app.py

@@ -112,7 +112,7 @@ def run():
             canvas.messages.append({"role": "user", "content": req["message"], "id": message_id})
             if len([m for m in canvas.messages if m["role"] == "user"]) > 1:
                 ten = TenantService.get_by_user_id(current_user.id)[0]
-                req["message"] = full_question(ten["tenant_id"], ten["llm_id"], canvas.messages)
+                #req["message"] = full_question(ten["tenant_id"], ten["llm_id"], canvas.messages)
             canvas.add_user_input(req["message"])
         answer = canvas.run(stream=stream)
         print(canvas)