A programming framework for agentic AI
Adding callable summary_method support and enhancements to initiate_chats (#1628)
Qingyun Wu899b250c08c1
diff --git a/autogen/agentchat/conversable_agent.py b/autogen/agentchat/conversable_agent.pyindex fbba53e5e..895eb7b51 100644--- a/autogen/agentchat/conversable_agent.py+++ b/autogen/agentchat/conversable_agent.py@@ -66,6 +66,7 @@ class ConversableAgent(LLMAgent):MAX_CONSECUTIVE_AUTO_REPLY = 100 # maximum number of consecutive auto replies (subject to future change)DEFAULT_summary_prompt = "Summarize the takeaway from the conversation. Do not add any introductory phrases."+ DEFAULT_summary_method = "last_msg"llm_config: Union[Dict, Literal[False]]def __init__(@@ -766,12 +767,20 @@ class ConversableAgent(LLMAgent):cache (Cache or None): the cache client to be used for this conversation.**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 specify the method to get a summary from the chat.- Supported methods 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" requires the llm_config to be set in either the sender or the recipient so that an llm client is available.- When both the sender and the recipient have an llm client, the recipient's llm client will be used.+ "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": a string of text used to prompt a LLM-based agent (the sender or receiver agent) to reflexton the conversation and extract a summary when summary_method is "reflection_with_llm".Default is DEFAULT_summary_prompt, i.e., "Summarize takeaway from the conversation. Do not add any introductory phrases. If the intended request is NOT properly addressed, please point it out."@@ -785,6 +794,9 @@ class ConversableAgent(LLMAgent):Returns:ChatResult: an ChatResult object."""+ _chat_info = context.copy()+ _chat_info["recipient"] = recipient+ self._consolidate_chat_info(_chat_info)for agent in [self, recipient]:agent._raise_exception_on_async_reply_functions()agent.previous_cache = agent.client_cache@@ -792,7 +804,7 @@ class ConversableAgent(LLMAgent):self._prepare_chat(recipient, clear_history)self.send(self.generate_init_message(**context), recipient, silent=silent)summary = self._summarize_chat(- context.get("summary_method"),+ context.get("summary_method", ConversableAgent.DEFAULT_summary_method),recipient,prompt=context.get("summary_prompt"),cache=cache,@@ -827,13 +839,16 @@ class ConversableAgent(LLMAgent):Returns:ChatResult: an ChatResult object."""+ _chat_info = context.copy()+ _chat_info["recipient"] = recipient+ self._consolidate_chat_info(_chat_info)self._prepare_chat(recipient, clear_history)for agent in [self, recipient]:agent.previous_cache = agent.client_cacheagent.client_cache = cacheawait self.a_send(await self.a_generate_init_message(**context), recipient, silent=silent)summary = self._summarize_chat(- context.get("summary_method"),+ context.get("summary_method", ConversableAgent.DEFAULT_summary_method),recipient,prompt=context.get("summary_prompt"),cache=cache,@@ -851,24 +866,34 @@ class ConversableAgent(LLMAgent):def _summarize_chat(self,- method,- agent: Optional[Agent] = None,+ summary_method,+ recipient: Optional[Agent] = None,prompt: Optional[str] = None,cache: Optional[Cache] = None,) -> str:"""Get a chat summary from an agent participating in a chat.Args:- method (str): the method to get the summary.- agent: the participating agent in a chat.+ summary_method (str or callable): the summary_method to get the summary.+ The 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"]+ ```+ recipient: the recipient agent in a chat.prompt (str): the prompt used to get a summary when summary_method is "reflection_with_llm".Returns:str: a chat summary from the agent."""- agent = self if agent is None else agent+ agent = self if recipient is None else recipientsummary = ""- if method == "reflection_with_llm":+ if summary_method is None:+ return summary+ if summary_method == "reflection_with_llm":prompt = ConversableAgent.DEFAULT_summary_prompt if prompt is None else promptif not isinstance(prompt, str):raise ValueError("The summary_prompt must be a string.")@@ -877,13 +902,13 @@ class ConversableAgent(LLMAgent):summary = self._reflection_with_llm(prompt, msg_list, llm_agent=agent, cache=cache)except BadRequestError as e:warnings.warn(f"Cannot extract summary using reflection_with_llm: {e}", UserWarning)- elif method == "last_msg" or method is None:+ elif summary_method == "last_msg" or summary_method is None:try:summary = agent.last_message(self)["content"].replace("TERMINATE", "")except (IndexError, AttributeError) as e:warnings.warn(f"Cannot extract summary using last_msg: {e}", UserWarning)- else:- warnings.warn(f"Unsupported summary method: {method}", UserWarning)+ elif isinstance(summary_method, Callable):+ summary = summary_method(recipient, self)return summarydef _reflection_with_llm(@@ -914,6 +939,22 @@ class ConversableAgent(LLMAgent):response = self._generate_oai_reply_from_client(llm_client=llm_client, messages=messages, cache=cache)return response+ def _consolidate_chat_info(self, chat_info: Union[Dict, List[Dict]]):+ if isinstance(chat_info, dict):+ chat_info = [chat_info]+ for c in chat_info:+ assert "recipient" in c, "recipient must be provided."+ summary_method = c.get("summary_method")+ assert (+ summary_method is None+ or isinstance(summary_method, Callable)+ or summary_method in ("last_msg", "reflection_with_llm")+ ), "summary_method must be a string chosen from 'reflection_with_llm' or 'last_msg' or a callable, or None."+ if summary_method == "reflection_with_llm":+ assert (+ self.client is not None or c["recipient"].client is not None+ ), "llm client must be set in either the recipient or sender when summary_method is reflection_with_llm."+def initiate_chats(self, chat_queue: List[Dict[str, Any]]) -> Dict[Agent, ChatResult]:"""(Experimental) Initiate chats with multiple agents.TODO: add async version of this method.@@ -925,12 +966,20 @@ class ConversableAgent(LLMAgent):- "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" can be used to specify the method to extract a summary from the chat.- Supported methods 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.+ "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."@@ -940,13 +989,17 @@ class ConversableAgent(LLMAgent):Returns: a dictionary of ChatResult object from the finished chats of particular agents."""+ self._consolidate_chat_info(chat_queue)receipts_set = set()for chat_info in chat_queue:assert "recipient" in chat_info, "recipient must be provided."receipts_set.add(chat_info["recipient"])- assert len(receipts_set) == len(chat_queue), "recipients must be different."-- self._chat_queue = chat_queue+ if len(receipts_set) < len(chat_queue):+ warnings.warn(+ "Repetitive recipients detected: The chat history will be cleared by default if a recipient appears more than once. To retain the chat history, please set 'clear_history=False' in the configuration of the repeating agent.",+ UserWarning,+ )+ self._chat_queue = chat_queue.copy()self._finished_chats = {}while self._chat_queue:chat_info = self._chat_queue.pop(0)@@ -1971,11 +2024,20 @@ class ConversableAgent(LLMAgent):Args:**context: any context information. It has the following reserved fields:"message": a str of message.- "summary_method": a string specify the method to get a summary from the chat.- Supported methods 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" requires the llm_config to be set in either the sender or the recipient so that an llm client is available.+ "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"]+ ```When both the sender and the recipient have an llm client, the recipient's llm client will be used."summary_prompt": a string of text used to prompt a LLM-based agent (the sender or receiver agent) to reflexton the conversation and extract a summary when summary_method is "reflection_with_llm".diff --git a/autogen/coding/embedded_ipython_code_executor.py b/autogen/coding/embedded_ipython_code_executor.pyindex 01a640015..c85798f75 100644--- a/autogen/coding/embedded_ipython_code_executor.py+++ b/autogen/coding/embedded_ipython_code_executor.py@@ -14,7 +14,7 @@ from ..agentchat.agent import LLMAgentfrom .base import CodeBlock, CodeExtractor, CodeResultfrom .markdown_code_extractor import MarkdownCodeExtractor-__all__ = ("EmbeddedIPythonCodeExecutor",)+__all__ = ("EmbeddedIPythonCodeExecutor", "IPythonCodeResult")class IPythonCodeResult(CodeResult):diff --git a/test/agentchat/test_chats.py b/test/agentchat/test_chats.pyindex 9b578c00c..dfa1617cb 100644--- a/test/agentchat/test_chats.py+++ b/test/agentchat/test_chats.py@@ -148,8 +148,8 @@ def test_chats():financial_tasks = ["""What are the full names of NVDA and TESLA.""",- """Investigate the reasons.""",- """Pros and cons of the companies I'm interested in. Keep it short.""",+ """Get their stock price.""",+ """Analyze pros and cons. Keep it short.""",]writing_tasks = ["""Develop a short but engaging blog post using any information provided."""]@@ -185,20 +185,29 @@ def test_chats():}, # Please set use_docker=True if docker is available to run the generated code. Using docker is safer than running the generated code directly.)+ def my_summary_method(recipient, sender):+ return recipient.chat_messages[sender][0].get("content", "")+chat_res = user.initiate_chats([{"recipient": financial_assistant_1,"message": financial_tasks[0],- "clear_history": True,"silent": False,- "summary_method": "last_msg",+ "summary_method": my_summary_method,},{"recipient": financial_assistant_2,"message": financial_tasks[1],+ "silent": True,"summary_method": "reflection_with_llm",},+ {+ "recipient": financial_assistant_1,+ "message": financial_tasks[2],+ "summary_method": "last_msg",+ "clear_history": False,+ },{"recipient": writer,"message": writing_tasks[0],@@ -216,9 +225,97 @@ def test_chats():print(writer_res.summary, writer_res.cost)print(all_res[financial_assistant_1].human_input)print(all_res[financial_assistant_1].summary)+ print(all_res[financial_assistant_1].chat_history)+ print(all_res[financial_assistant_2].summary)# print(blogpost.summary, insights_and_blogpost)[email protected](skip_openai, reason="requested to skip openai tests")+def test_chats_exceptions():+ config_list = autogen.config_list_from_json(+ OAI_CONFIG_LIST,+ file_location=KEY_LOC,+ )++ financial_tasks = [+ """What are the full names of NVDA and TESLA.""",+ """Get their stock price.""",+ """Analyze pros and cons. Keep it short.""",+ ]++ financial_assistant_1 = AssistantAgent(+ name="Financial_assistant_1",+ llm_config={"config_list": config_list},+ )+ financial_assistant_2 = AssistantAgent(+ name="Financial_assistant_2",+ llm_config={"config_list": config_list},+ )+ user = UserProxyAgent(+ name="User",+ human_input_mode="NEVER",+ is_termination_msg=lambda x: x.get("content", "").find("TERMINATE") >= 0,+ code_execution_config={+ "last_n_messages": 1,+ "work_dir": "tasks",+ "use_docker": False,+ }, # Please set use_docker=True if docker is available to run the generated code. Using docker is safer than running the generated code directly.+ )++ user_2 = UserProxyAgent(+ name="User",+ human_input_mode="NEVER",+ is_termination_msg=lambda x: x.get("content", "").find("TERMINATE") >= 0,+ code_execution_config={+ "last_n_messages": 1,+ "work_dir": "tasks",+ "use_docker": False,+ }, # Please set use_docker=True if docker is available to run the generated code. Using docker is safer than running the generated code directly.+ )++ with pytest.raises(+ AssertionError,+ match="summary_method must be a string chosen from 'reflection_with_llm' or 'last_msg' or a callable, or None.",+ ):+ user.initiate_chats(+ [+ {+ "recipient": financial_assistant_1,+ "message": financial_tasks[0],+ "silent": False,+ "summary_method": "last_msg",+ },+ {+ "recipient": financial_assistant_2,+ "message": financial_tasks[2],+ "summary_method": "llm",+ "clear_history": False,+ },+ ]+ )++ with pytest.raises(+ AssertionError,+ match="llm client must be set in either the recipient or sender when summary_method is reflection_with_llm.",+ ):+ user.initiate_chats(+ [+ {+ "recipient": financial_assistant_1,+ "message": financial_tasks[0],+ "silent": False,+ "summary_method": "last_msg",+ },+ {+ "recipient": user_2,+ "message": financial_tasks[2],+ "clear_history": False,+ "summary_method": "reflection_with_llm",+ },+ ]+ )++@pytest.mark.skipif(skip_openai, reason="requested to skip openai tests")def test_chats_w_func():config_list = autogen.config_list_from_json(@@ -281,7 +378,8 @@ def test_chats_w_func():if __name__ == "__main__":- # test_chats()+ test_chats()+ # test_chats_exceptions()# test_chats_group()# test_chats_w_func()- test_chat_messages_for_summary()+ # test_chat_messages_for_summary()diff --git a/test/agentchat/test_function_call.py b/test/agentchat/test_function_call.pyindex 52430c70b..09d17db5e 100644--- a/test/agentchat/test_function_call.py+++ b/test/agentchat/test_function_call.py@@ -256,10 +256,30 @@ def test_update_function():assert "greet_user" not in messages2print("Chat summary and cost", res2.summary, res2.cost)+ with pytest.raises(+ AssertionError,+ match="summary_method must be a string chosen from 'reflection_with_llm' or 'last_msg' or a callable, or None.",+ ):+ user_proxy.initiate_chat(+ assistant,+ message="What functions do you know about in the context of this conversation? End your response with 'TERMINATE'.",+ summary_method="llm",+ )++ with pytest.raises(+ AssertionError,+ match="llm client must be set in either the recipient or sender when summary_method is reflection_with_llm.",+ ):+ user_proxy.initiate_chat(+ recipient=user_proxy,+ message="What functions do you know about in the context of this conversation? End your response with 'TERMINATE'.",+ summary_method="reflection_with_llm",+ )+if __name__ == "__main__":# test_json_extraction()# test_execute_function()test_update_function()- asyncio.run(test_a_execute_function())- test_eval_math_responses()+ # asyncio.run(test_a_execute_function())+ # test_eval_math_responses()diff --git a/test/coding/test_embedded_ipython_code_executor.py b/test/coding/test_embedded_ipython_code_executor.pyindex d16690960..3ecb318f9 100644--- a/test/coding/test_embedded_ipython_code_executor.py+++ b/test/coding/test_embedded_ipython_code_executor.py@@ -93,7 +93,7 @@ def test_timeout() -> None:@pytest.mark.skipif(skip, reason=skip_reason)def test_silent_pip_install() -> None:- executor = EmbeddedIPythonCodeExecutor()+ executor = EmbeddedIPythonCodeExecutor(timeout=600)code_blocks = [CodeBlock(code="!pip install matplotlib numpy", language="python")]code_result = executor.execute_code_blocks(code_blocks)assert code_result.exit_code == 0 and code_result.output.strip() == ""