Merge pull request #123 from alipay/dev

feat: Version 0.0.11 Release

Author: LandJerry

CHANGELOG.md

@@ -24,6 +24,19 @@ Note - Additional remarks regarding the version.
 ***************************************************
 
 # Version Update History
+## [0.0.11] - 2024-07-11
+### Added
+- DataAgent Autonomous Data Agent MVP Version Released
+  - Minimum Viable Product version, DataAgent aims to empower your agent with the capability of self-assessment and evolution through intelligent agent abilities. For detailed information, please refer to the user documentation.
+- Added intermediate information streaming output capabilities in PEER and ReAct modes
+
+### Note
+- Latest PEER research findings released
+  - This paper provides a detailed introduction to the mechanisms and principles of the PEER multi-agent framework. Experimental validation proves the advancement of the PEER model. For detailed information, please refer to the user documentation.
+- Added use cases
+  - Andrew Ng's Reflexive Workflow Translation Agent Replication
+- Some code optimizations and documentation updates.
+
 ## [0.0.10] - 2024-06-28
 ### Added
 - Added standard integration for the DeepSeek model in the LLM module.

CHANGELOG_zh.md

@@ -24,6 +24,19 @@ Note - 对于版本的额外说明。
 ***************************************************
 
 # 版本更新记录
+## [0.0.11] - 2024-07-11
+### Added
+- DataAgent数据自治智能体MVP版本发布
+  - Minimum Viable Product版本，DataAgent旨在使用智能体能力让您的Agent拥有自我评价与演进的能力，详细内容请查阅用户文档。
+- 增加PEER、ReAct模式中间信息流式输出能力
+
+### Note
+- PEER最新研究成果发布
+  - 该文献详细介绍了介绍了PEER多智能体框架的机制原理，同时在实验部分通过验证证明了PEER模式的先进性，详细内容请查阅用户文档。
+- 新增使用案例
+  - 吴恩达反思工作流翻译智能体复刻
+- 部分代码优化与文档更新
+
 ## [0.0.10] - 2024-06-28
 ### Added
 - LLM组件新增DeepSeek模型标准接入

README.md

@@ -5,7 +5,7 @@ Language version: [English](./README.md) | [中文](./README_zh.md) | [日本語
 ![](https://img.shields.io/badge/framework-agentUniverse-pink)
 ![](https://img.shields.io/badge/python-3.10%2B-blue?logo=Python)
 [![](https://img.shields.io/badge/%20license-Apache--2.0-yellow)](LICENSE)
-[![Static Badge](https://img.shields.io/badge/pypi-v0.0.10-blue?logo=pypi)](https://pypi.org/project/agentUniverse/)
+[![Static Badge](https://img.shields.io/badge/pypi-v0.0.11-blue?logo=pypi)](https://pypi.org/project/agentUniverse/)
 
 ![](docs/guidebook/_picture/logo_bar.jpg)
 ****************************************
@@ -35,6 +35,7 @@ More patterns are coming soon...
   * [User Guide](#User-Guide)
   * [API Reference](#API-Reference)
   * [Support](#Support)
+  * [Citation](#Citation)
   * [Acknowledgements](#Acknowledgements)
 ****************************************
 ## Quick Start
@@ -64,6 +65,11 @@ For more details, please read the [Quick Start](./docs/guidebook/en/1_3_Quick_St
 
 [Financial Event Analysis Based on PEER Multi-Agent Mode](./docs/guidebook/en/6_4_1_Financial_Event_Analysis_Case.md)
 
+[Andrew Ng's Reflexive Workflow Translation Agent Replication](./docs/guidebook/en/7_1_1_Translation_Case.md)
+
+#### 🚩 DataAgent - Data Autonomous Agent
+agentUniverse has launched DataAgent (Minimum Viable Product Version). DataAgent aims to empower your agent with the capability of self-assessment and evolution through the use of intelligent agent abilities. For more details, please refer to the documentation. [DataAgent - Data Autonomous Agent](./docs/guidebook/en/8_1_1_data_autonomous_agent.md)
+
 ### 🌟 Example Projects
 [agentUniverse Example Projects](sample_standard_app)
 
@@ -101,11 +107,26 @@ For more details, please read the [Quick Start](./docs/guidebook/en/1_3_Quick_St
 [jerry.zzw@antgroup.com](mailto:jerry.zzw@antgroup.com)
 [jinshi.zjs@antgroup.com](mailto:jinshi.zjs@antgroup.com)
 
-#### WeChat Official Account
-
-😊 Official Account ID: **agentUniverse智多星** 
-
-You can get more related articles and information in our WeChat Official Account.
+#### twitter
+ID: [@agentuniverse_](https://x.com/agentuniverse_)
+
+### Citation
+The agentUniverse project is supported by the following research achievements.
+
+BibTeX formatted
+```text
+@misc{wang2024peerexpertizingdomainspecifictasks,
+      title={PEER: Expertizing Domain-Specific Tasks with a Multi-Agent Framework and Tuning Methods}, 
+      author={Yiying Wang and Xiaojing Li and Binzhu Wang and Yueyang Zhou and Han Ji and Hong Chen and Jinshi Zhang and Fei Yu and Zewei Zhao and Song Jin and Renji Gong and Wanqing Xu},
+      year={2024},
+      eprint={2407.06985},
+      archivePrefix={arXiv},
+      primaryClass={cs.AI},
+      url={https://arxiv.org/abs/2407.06985}, 
+}
+```
+Overview: This document provides a detailed introduction to the mechanisms and principles of the PEER multi-agent framework. In the experimental section, scores were assigned across seven dimensions: completeness, relevance, conciseness, factualness, logicality, structure, and comprehensiveness (each dimension has a maximum score of 5 points). The PEER model scored higher on average in each evaluation dimension compared to BabyAGI and demonstrated significant advantages in the dimensions of completeness, relevance, logicality, structure, and comprehensiveness. Additionally, the PEER model achieved a superior rate of 83% over BabyAGI using the GPT-3.5 Turbo (16k) model, and 81% using the GPT-4 model. For more details, please refer to the document.
+https://arxiv.org/pdf/2407.06985
 
 ## Acknowledgements
 This project is partially built on excellent open-source projects such as langchain, pydantic, gunicorn, flask, SQLAlchemy, chromadb, etc. (The detailed dependency list can be found in pyproject.toml). We would like to extend special thanks to the related projects and contributors. 🙏🙏🙏

README_zh.md

@@ -5,7 +5,7 @@
 ![](https://img.shields.io/badge/framework-agentUniverse-pink)
 ![](https://img.shields.io/badge/python-3.10%2B-blue?logo=Python)
 [![](https://img.shields.io/badge/%20license-Apache--2.0-yellow)](LICENSE)
-[![Static Badge](https://img.shields.io/badge/pypi-v0.0.10-blue?logo=pypi)](https://pypi.org/project/agentUniverse/)
+[![Static Badge](https://img.shields.io/badge/pypi-v0.0.11-blue?logo=pypi)](https://pypi.org/project/agentUniverse/)
 
 ![](docs/guidebook/_picture/logo_bar.jpg)
 ****************************************
@@ -34,6 +34,7 @@
   * [用户指南](#用户指南)
   * [API参考](#API参考)
   * [支持](#支持)
+  * [文献](#文献)
   * [鸣谢](#鸣谢)
 ****************************************
 ## 快速开始
@@ -61,6 +62,12 @@ pip install agentUniverse
 
 [基于PEER协同模式的金融事件分析](./docs/guidebook/zh/6_4_1_金融事件分析案例.md)
 
+[吴恩达反思工作流翻译智能体复刻](./docs/guidebook/zh/7_1_1_翻译案例.md)
+
+
+#### 🚩 DataAgent - 数据自治智能体
+agentUniverse推出了DataAgent（Minimum Viable Product版本）, DataAgent旨在使用智能体能力让您的Agent拥有自我评价与演进的能力。详情见文档: [DataAgent - 数据自治智能体](./docs/guidebook/zh/8_1_1_数据自治智能体.md)
+
 ### 🌟 示例项目
 [agentUniverse 示例项目](sample_standard_app)
 
@@ -105,7 +112,30 @@ pip install agentUniverse
 
 😊 公众号ID：**agentUniverse智多星**
 
+![](./docs/guidebook/_picture/wechat_official.png)
+
 更多相关的文章与资讯你可以在微信公众号中获取。
 
+#### twitter
+ID: [@agentuniverse_](https://x.com/agentuniverse_)
+
+### 文献
+agentUniverse项目基于以下的研究成果支撑。
+
+BibTeX formatted
+```text
+@misc{wang2024peerexpertizingdomainspecifictasks,
+      title={PEER: Expertizing Domain-Specific Tasks with a Multi-Agent Framework and Tuning Methods}, 
+      author={Yiying Wang and Xiaojing Li and Binzhu Wang and Yueyang Zhou and Han Ji and Hong Chen and Jinshi Zhang and Fei Yu and Zewei Zhao and Song Jin and Renji Gong and Wanqing Xu},
+      year={2024},
+      eprint={2407.06985},
+      archivePrefix={arXiv},
+      primaryClass={cs.AI},
+      url={https://arxiv.org/abs/2407.06985}, 
+}
+```
+文献简介：该文献详细介绍了介绍了PEER多智能体框架的机制原理，同时在实验部分分别从**完整性、相关性、紧凑性、事实性、逻辑性、结构性和全面性七个维度进行打分（各纬度满分为5分）**，PEER模式在每个测评维度的平均分数均高于BabyAGI，且在**完整性、相关性、逻辑性、结构性和全面性五个纬度有显著优势**；同时PEER模式在 GPT-3.5 turbo (16k) 模型下相较于 BabyAGI 的择优胜率达到 83%，在 GPT-4o 模型下择优胜率达到 81%，更多详情请阅读文献。
+https://arxiv.org/pdf/2407.06985
+
 ## 鸣谢
 本项目部分基于langchain、pydantic、gunicorn、flask、SQLAlchemy、chromadb等（详细依赖列表可见pyproject.toml）优秀开源项目实现，在此特别感谢相关项目与关联方。 🙏🙏🙏

agentuniverse/agent/default/executing_agent/executing_agent.py

@@ -76,7 +76,8 @@ def execute(self, input_object: InputObject, agent_input: dict) -> dict:
         framework = agent_input.get('framework', [])
         futures = []
         for task in framework:
-            agent_input_copy: dict = copy.deepcopy(agent_input)
+            # note: agent input shallow copy.
+            agent_input_copy: dict = dict(agent_input)
             agent_input_copy['input'] = task
             planner: Planner = PlannerManager().get_instance_obj(self.agent_model.plan.get('planner').get('name'))
             futures.append(
@@ -99,7 +100,8 @@ def process_intput_object(self, input_object: InputObject, subtask: str, planner
         # get agent toolsets.
         action: dict = self.agent_model.action or dict()
         tools: list = action.get('tool') or list()
-        input_object_copy: InputObject = copy.deepcopy(input_object)
+        # note: input object shallow copy.
+        input_object_copy: InputObject = InputObject(input_object.to_dict())
         # wrap input_object for agent knowledge.
         input_object_copy.add_data(planner_input_key, subtask)
         # wrap input_object for agent toolsets.

agentuniverse/agent/memory/chat_memory.py

@@ -50,17 +50,18 @@ def as_langchain(self) -> BaseChatMemory:
                                                      max_token_limit=self.max_tokens, messages=self.messages,
                                                      prompt_version=self.prompt_version)
 
-    def set_by_agent_model(self, **kwargs) -> None:
+    def set_by_agent_model(self, **kwargs):
         """ Assign values of parameters to the ChatMemory model in the agent configuration."""
-        super().set_by_agent_model(**kwargs)
+        copied_obj = super().set_by_agent_model(**kwargs)
         if 'messages' in kwargs and kwargs['messages']:
-            self.messages = kwargs['messages']
+            copied_obj.messages = kwargs['messages']
         if 'llm' in kwargs and kwargs['llm']:
-            self.llm = kwargs['llm']
+            copied_obj.llm = kwargs['llm']
         if 'input_key' in kwargs and kwargs['input_key']:
-            self.input_key = kwargs['input_key']
+            copied_obj.input_key = kwargs['input_key']
         if 'output_key' in kwargs and kwargs['output_key']:
-            self.output_key = kwargs['output_key']
+            copied_obj.output_key = kwargs['output_key']
+        return copied_obj
 
     def initialize_by_component_configer(self, component_configer: MemoryConfiger) -> 'ChatMemory':
         """Initialize the chat memory by the ComponentConfiger object.

agentuniverse/agent/memory/memory.py

@@ -40,12 +40,15 @@ def as_langchain(self) -> BaseMemory:
         """Convert the agentUniverse(aU) memory class to the langchain memory class."""
         pass
 
-    def set_by_agent_model(self, **kwargs) -> None:
+    def set_by_agent_model(self, **kwargs):
         """ Assign values of parameters to the Memory model in the agent configuration."""
+        # note: default shallow copy
+        copied_obj = self.model_copy()
         if 'memory_key' in kwargs and kwargs['memory_key']:
-            self.memory_key = kwargs['memory_key']
+            copied_obj.memory_key = kwargs['memory_key']
         if 'max_tokens' in kwargs and kwargs['max_tokens']:
-            self.max_tokens = kwargs['max_tokens']
+            copied_obj.max_tokens = kwargs['max_tokens']
+        return copied_obj
 
     def get_instance_code(self) -> str:
         """Return the full name of the memory."""

agentuniverse/agent/plan/planner/expressing_planner/expressing_planner.py

@@ -51,8 +51,8 @@ def invoke(self, agent_model: AgentModel, planner_input: dict, input_object: Inp
             history_messages_key="chat_history",
             input_messages_key=self.input_key,
         ) | StrOutputParser()
-        res = asyncio.run(
-            chain_with_history.ainvoke(input=planner_input, config={"configurable": {"session_id": "unused"}}))
+
+        res = self.invoke_chain(agent_model, chain_with_history, planner_input, chat_history, input_object)
         return {**planner_input, self.output_key: res, 'chat_history': generate_memories(chat_history)}
 
     def handle_prompt(self, agent_model: AgentModel, planner_input: dict) -> Prompt:

agentuniverse/agent/plan/planner/peer_planner/peer_planner.py

@@ -42,7 +42,7 @@ def invoke(self, agent_model: AgentModel, planner_input: dict, input_object: Inp
         """
         planner_config = agent_model.plan.get('planner')
         sub_agents = self.generate_sub_agents(planner_config)
-        return self.agents_run(sub_agents, planner_config, planner_input, input_object)
+        return self.agents_run(agent_model, sub_agents, planner_config, planner_input, input_object)
 
     @staticmethod
     def generate_sub_agents(planner_config: dict) -> dict:
@@ -79,7 +79,8 @@ def build_expert_framework(planner_config: dict, input_object: InputObject):
             elif context:
                 input_object.add_data('expert_framework', context)
 
-    def agents_run(self, agents: dict, planner_config: dict, agent_input: dict, input_object: InputObject) -> dict:
+    def agents_run(self, agent_mode: AgentModel, agents: dict, planner_config: dict, agent_input: dict,
+                   input_object: InputObject) -> dict:
         """Planner agents run.
 
         Args:
@@ -125,6 +126,10 @@ def agents_run(self, agents: dict, planner_config: dict, agent_input: dict, inpu
                 for index, one_framework in enumerate(planning_result.get_data('framework')):
                     logger_info += f"[{index + 1}] {one_framework} \n"
                 LOGGER.info(logger_info)
+                self.stream_output(input_object, {"data": {
+                    'output': planning_result.to_dict(),
+                    "agent_info": agent_mode.info
+                }, "type": "planning"})
 
             if not executing_result or jump_step in ["planning", "executing"]:
                 if not executingAgent:
@@ -143,6 +148,10 @@ def agents_run(self, agents: dict, planner_config: dict, agent_input: dict, inpu
                         one_exec_log_info += f"[{index + 1}] output: {one_exec_res['output']}\n"
                         logger_info += one_exec_log_info
                 LOGGER.info(logger_info)
+                self.stream_output(input_object, {"data": {
+                    'output': executing_result.to_dict(),
+                    "agent_info": agent_mode.info
+                }, "type": "executing"})
 
             if not expressing_result or jump_step in ["planning", "executing", "expressing"]:
                 if not expressingAgent:
@@ -157,6 +166,10 @@ def agents_run(self, agents: dict, planner_config: dict, agent_input: dict, inpu
                 logger_info = f"\nExpressing agent execution result is :\n"
                 logger_info += f"{expressing_result.get_data('output')}"
                 LOGGER.info(logger_info)
+                self.stream_output(input_object, {"data": {
+                    'output': expressing_result.get_data('output'),
+                    "agent_info": agent_mode.info
+                }, "type": "expressing"})
 
             if not reviewing_result or jump_step in ["planning", "executing", "expressing", "reviewing"]:
                 if not reviewingAgent:
@@ -172,14 +185,15 @@ def agents_run(self, agents: dict, planner_config: dict, agent_input: dict, inpu
                 else:
                     LOGGER.info(f"Starting reviewing agent.")
                     reviewing_result = reviewingAgent.run(**input_object.to_dict())
+
                     input_object.add_data('reviewing_result', reviewing_result)
 
                     # add reviewing agent log info
                     logger_info = f"\nReviewing agent execution result is :\n"
                     reviewing_info_str = f"review suggestion: {reviewing_result.get_data('suggestion')} \n"
                     reviewing_info_str += f"review score: {reviewing_result.get_data('score')} \n"
                     LOGGER.info(logger_info + reviewing_info_str)
-
+                    self.stream_output(input_object, {"data": reviewing_result.to_dict(), "type": "reviewing"})
                     if reviewing_result.get_data('score') and reviewing_result.get_data('score') >= eval_threshold:
                         loopResults.append({
                             "planning_result": planning_result,