Allow limiting the maximum number of turns in initiate_chat and `in…

…itiate_chats` (#1703) * max_turns * add notebook * simplify code * groupchat _prepare_chat * doc * add doc * Update autogen/agentchat/conversable_agent.py --------- Co-authored-by: Chi Wang <[email protected]>
microsoft · Feb 17, 2024 · a00df55 · a00df55
1 parent bf3cc0f
commit a00df55
Show file tree

Hide file tree

Showing 10 changed files with 1,106 additions and 464 deletions.
diff --git a/autogen/agentchat/chat.py b/autogen/agentchat/chat.py
@@ -34,31 +34,37 @@ def initiate_chats(chat_queue: List[Dict[str, Any]]) -> List[ChatResult]:
 
     args:
         chat_queue (List[Dict]): a list of dictionaries containing the information of the chats.
-                Each dictionary should contain the following fields:
+                Each dictionary should contain the input arguments for `ConversableAgent.initiate_chat`.
+                More specifically, each dictionary could include the following fields:
+                recipient: the recipient agent.
+                - "sender": the sender agent.
                 - "recipient": the recipient agent.
-                - "context": any context information, e.g., the request message. The following fields are reserved:
-                    "message" needs to be provided if the `generate_init_message` method is not overridden.
-                          Otherwise, input() will be called to get the initial message.
-                    "summary_method": a string or callable specifying the method to get a summary from the chat. Default is DEFAULT_summary_method, i.e., "last_msg".
-                        - Supported string are "last_msg" and "reflection_with_llm":
-                            when set "last_msg", it returns the last message of the dialog as the summary.
-                            when set "reflection_with_llm", it returns a summary extracted using an llm client.
-                            `llm_config` must be set in either the recipient or sender.
-                            "reflection_with_llm" requires the llm_config to be set in either the sender or the recipient.
-                        - A callable summary_method should take the recipient and sender agent in a chat as input and return a string of summary. E.g,
-                        ```python
-                        def my_summary_method(
-                            sender: ConversableAgent,
-                            recipient: ConversableAgent,
-                        ):
-                            return recipient.last_message(sender)["content"]
-                        ```
-                    "summary_prompt" can be used to specify the prompt used to extract a summary when summary_method is "reflection_with_llm".
-                        Default is None and the following default prompt will be used when "summary_method" is set to "reflection_with_llm":
-                        "Identify and extract the final solution to the originally asked question based on the conversation."
-                    "carryover" can be used to specify the carryover information to be passed to this chat.
-                        If provided, we will combine this carryover with the "message" content when generating the initial chat
-                        message in `generate_init_message`.
+                - clear_history (bool): whether to clear the chat history with the agent. Default is True.
+                - silent (bool or None): (Experimental) whether to print the messages for this conversation. Default is False.
+                - cache (Cache or None): the cache client to be used for this conversation. Default is None.
+                - max_turns (int or None): the maximum number of turns for the chat. If None, the chat will continue until a termination condition is met. Default is None.
+                - "message" needs to be provided if the `generate_init_message` method is not overridden.
+                        Otherwise, input() will be called to get the initial message.
+                - "summary_method": a string or callable specifying the method to get a summary from the chat. Default is DEFAULT_summary_method, i.e., "last_msg".
+                    - Supported string are "last_msg" and "reflection_with_llm":
+                        when set "last_msg", it returns the last message of the dialog as the summary.
+                        when set "reflection_with_llm", it returns a summary extracted using an llm client.
+                        `llm_config` must be set in either the recipient or sender.
+                        "reflection_with_llm" requires the llm_config to be set in either the sender or the recipient.
+                    - A callable summary_method should take the recipient and sender agent in a chat as input and return a string of summary. E.g,
+                    ```python
+                    def my_summary_method(
+                        sender: ConversableAgent,
+                        recipient: ConversableAgent,
+                    ):
+                        return recipient.last_message(sender)["content"]
+                    ```
+                "summary_prompt" can be used to specify the prompt used to extract a summary when summary_method is "reflection_with_llm".
+                    Default is None and the following default prompt will be used when "summary_method" is set to "reflection_with_llm":
+                    "Identify and extract the final solution to the originally asked question based on the conversation."
+                "carryover" can be used to specify the carryover information to be passed to this chat.
+                    If provided, we will combine this carryover with the "message" content when generating the initial chat
+                    message in `generate_init_message`.
 
 
     returns:

diff --git a/autogen/agentchat/conversable_agent.py b/autogen/agentchat/conversable_agent.py
@@ -519,13 +519,6 @@ def send(
                 "Message can't be converted into a valid ChatCompletion message. Either content or function_call must be provided."
             )
 
-        chat_result = ChatResult(
-            chat_history=self.chat_messages[recipient],
-            cost=gather_usage_summary([self, recipient]),
-            human_input=self._human_input,
-        )
-        return chat_result
-
     async def a_send(
         self,
         message: Union[Dict, str],
@@ -578,13 +571,6 @@ async def a_send(
                 "Message can't be converted into a valid ChatCompletion message. Either content or function_call must be provided."
             )
 
-        chat_result = ChatResult(
-            chat_history=self.chat_messages[recipient],
-            cost=gather_usage_summary([self, recipient]),
-            human_input=self._human_input,
-        )
-        return chat_result
-
     def _print_received_message(self, message: Union[Dict, str], sender: Agent):
         # print the message received
         print(colored(sender.name, "yellow"), "(to", f"{self.name}):\n", flush=True)
@@ -729,14 +715,20 @@ async def a_receive(
         if reply is not None:
             await self.a_send(reply, sender, silent=silent)
 
-    def _prepare_chat(self, recipient: "ConversableAgent", clear_history: bool, prepare_recipient: bool = True) -> None:
+    def _prepare_chat(
+        self,
+        recipient: "ConversableAgent",
+        clear_history: bool,
+        prepare_recipient: bool = True,
+        reply_at_receive: bool = True,
+    ) -> None:
         self.reset_consecutive_auto_reply_counter(recipient)
-        self.reply_at_receive[recipient] = True
+        self.reply_at_receive[recipient] = reply_at_receive
         if clear_history:
             self.clear_history(recipient)
             self._human_input = []
         if prepare_recipient:
-            recipient._prepare_chat(self, clear_history, False)
+            recipient._prepare_chat(self, clear_history, False, reply_at_receive)
 
     def _raise_exception_on_async_reply_functions(self) -> None:
         """Raise an exception if any async reply functions are registered.
@@ -763,6 +755,7 @@ def initiate_chat(
         clear_history: Optional[bool] = True,
         silent: Optional[bool] = False,
         cache: Optional[Cache] = None,
+        max_turns: Optional[int] = None,
         **context,
     ) -> ChatResult:
         """Initiate a chat with the recipient agent.
@@ -773,9 +766,12 @@ def initiate_chat(
 
         Args:
             recipient: the recipient agent.
-            clear_history (bool): whether to clear the chat history with the agent.
-            silent (bool or None): (Experimental) whether to print the messages for this conversation.
-            cache (Cache or None): the cache client to be used for this conversation.
+            clear_history (bool): whether to clear the chat history with the agent. Default is True.
+            silent (bool or None): (Experimental) whether to print the messages for this conversation. Default is False.
+            cache (Cache or None): the cache client to be used for this conversation. Default is None.
+            max_turns (int or None): the maximum number of turns for the chat between the two agents. One turn means one conversation round trip. Note that this is different from
+            [max_consecutive_auto_reply](#max_consecutive_auto_reply) which is the maximum number of consecutive auto replies; and it is also different from [max_rounds in GroupChat](./groupchat#groupchat-objects) which is the maximum number of rounds in a group chat session.
+            If max_turns is set to None, the chat will continue until a termination condition is met. Default is None.
             **context: any context information. It has the following reserved fields:
                 "message": a str of message. Needs to be provided. Otherwise, input() will be called to get the initial message.
                 "summary_method": a string or callable specifying the method to get a summary from the chat. Default is DEFAULT_summary_method, i.e., "last_msg".
@@ -812,8 +808,19 @@ def my_summary_method(
             agent._raise_exception_on_async_reply_functions()
             agent.previous_cache = agent.client_cache
             agent.client_cache = cache
-        self._prepare_chat(recipient, clear_history)
-        self.send(self.generate_init_message(**context), recipient, silent=silent)
+        if isinstance(max_turns, int):
+            self._prepare_chat(recipient, clear_history, reply_at_receive=False)
+            for _ in range(max_turns):
+                if _ == 0:
+                    msg2send = self.generate_init_message(**context)
+                else:
+                    msg2send = self.generate_reply(messages=self.chat_messages[recipient], sender=recipient)
+                if msg2send is None:
+                    break
+                self.send(msg2send, recipient, request_reply=True, silent=silent)
+        else:
+            self._prepare_chat(recipient, clear_history)
+            self.send(self.generate_init_message(**context), recipient, silent=silent)
         summary = self._summarize_chat(
             context.get("summary_method", ConversableAgent.DEFAULT_summary_method),
             recipient,
@@ -837,6 +844,7 @@ async def a_initiate_chat(
         clear_history: Optional[bool] = True,
         silent: Optional[bool] = False,
         cache: Optional[Cache] = None,
+        max_turns: Optional[int] = None,
         **context,
     ) -> ChatResult:
         """(async) Initiate a chat with the recipient agent.
@@ -853,11 +861,22 @@ async def a_initiate_chat(
         _chat_info = context.copy()
         _chat_info["recipient"] = recipient
         consolidate_chat_info(_chat_info, uniform_sender=self)
-        self._prepare_chat(recipient, clear_history)
         for agent in [self, recipient]:
             agent.previous_cache = agent.client_cache
             agent.client_cache = cache
-        await self.a_send(await self.a_generate_init_message(**context), recipient, silent=silent)
+        if isinstance(max_turns, int):
+            self._prepare_chat(recipient, clear_history, reply_at_receive=False)
+            for _ in range(max_turns):
+                if _ == 0:
+                    msg2send = await self.a_generate_init_message(**context)
+                else:
+                    msg2send = await self.a_generate_reply(messages=self.chat_messages[recipient], sender=recipient)
+                if msg2send is None:
+                    break
+                await self.a_send(msg2send, recipient, request_reply=True, silent=silent)
+        else:
+            self._prepare_chat(recipient, clear_history)
+            await self.a_send(await self.a_generate_init_message(**context), recipient, silent=silent)
         summary = self._summarize_chat(
             context.get("summary_method", ConversableAgent.DEFAULT_summary_method),
             recipient,
@@ -956,31 +975,7 @@ def initiate_chats(self, chat_queue: List[Dict[str, Any]]) -> List[ChatResult]:
 
         Args:
             chat_queue (List[Dict]): a list of dictionaries containing the information of the chats.
-                Each dictionary should contain the following fields:
-                - "recipient": the recipient agent.
-                - "context": any context information, e.g., the request message. The following fields are reserved:
-                    "message" needs to be provided if the `generate_init_message` method is not overridden.
-                          Otherwise, input() will be called to get the initial message.
-                    "summary_method": a string or callable specifying the method to get a summary from the chat. Default is DEFAULT_summary_method, i.e., "last_msg".
-                        - Supported string are "last_msg" and "reflection_with_llm":
-                            when set "last_msg", it returns the last message of the dialog as the summary.
-                            when set "reflection_with_llm", it returns a summary extracted using an llm client.
-                            `llm_config` must be set in either the recipient or sender.
-                            "reflection_with_llm" requires the llm_config to be set in either the sender or the recipient.
-                        - A callable summary_method should take the recipient and sender agent in a chat as input and return a string of summary. E.g,
-                        ```python
-                        def my_summary_method(
-                            sender: ConversableAgent,
-                            recipient: ConversableAgent,
-                        ):
-                            return recipient.last_message(sender)["content"]
-                        ```
-                    "summary_prompt" can be used to specify the prompt used to extract a summary when summary_method is "reflection_with_llm".
-                        Default is None and the following default prompt will be used when "summary_method" is set to "reflection_with_llm":
-                        "Identify and extract the final solution to the originally asked question based on the conversation."
-                    "carryover" can be used to specify the carryover information to be passed to this chat.
-                        If provided, we will combine this carryover with the "message" content when generating the initial chat
-                        message in `generate_init_message`.
+                Each dictionary should contain the input arguments for [`initiate_chat`](conversable_agent#initiate_chat)
 
         Returns: a list of ChatResult objects corresponding to the finished chats in the chat_queue.
         """

diff --git a/autogen/agentchat/groupchat.py b/autogen/agentchat/groupchat.py
@@ -507,15 +507,21 @@ def chat_messages_for_summary(self, agent: Agent) -> List[Dict]:
         """
         return self._groupchat.messages
 
-    def _prepare_chat(self, recipient: ConversableAgent, clear_history: bool, prepare_recipient: bool = True) -> None:
-        super()._prepare_chat(recipient, clear_history, prepare_recipient)
+    def _prepare_chat(
+        self,
+        recipient: ConversableAgent,
+        clear_history: bool,
+        prepare_recipient: bool = True,
+        reply_at_receive: bool = True,
+    ) -> None:
+        super()._prepare_chat(recipient, clear_history, prepare_recipient, reply_at_receive)
 
         if clear_history:
             self._groupchat.reset()
 
         for agent in self._groupchat.agents:
             if (recipient != agent or prepare_recipient) and isinstance(agent, ConversableAgent):
-                agent._prepare_chat(self, clear_history, False)
+                agent._prepare_chat(self, clear_history, False, reply_at_receive)
 
     def run_chat(
         self,