From 85e34807a259adb63afe96c937627b7a5df0577f Mon Sep 17 00:00:00 2001 From: zilun Date: Thu, 5 Dec 2024 19:06:58 +0800 Subject: [PATCH] this is a branch for COT --- examples/cotprompt/README.md | 47 +++++++++++ .../cotprompt/agent/cot_agent/__init__.py | 0 .../cotprompt/agent/cot_agent/cot_conclude.py | 57 +++++++++++++ .../agent/cot_agent/cot_input_interface.py | 21 +++++ .../agent/cot_agent/cot_reasoning.py | 64 ++++++++++++++ examples/cotprompt/compile_container.py | 17 ++++ examples/cotprompt/configs/llms/gpt4o.yml | 7 ++ examples/cotprompt/configs/llms/json_res.yml | 6 ++ examples/cotprompt/configs/llms/text_res.yml | 6 ++ .../cotprompt/configs/tools/all_tools.yml | 10 +++ .../configs/workers/cot_conclude.yml | 5 ++ .../configs/workers/cot_input_interface.yml | 5 ++ .../configs/workers/cot_reasoning.yaml | 5 ++ examples/cotprompt/container.yaml | 84 +++++++++++++++++++ examples/cotprompt/run_cli.py | 56 +++++++++++++ 15 files changed, 390 insertions(+) create mode 100644 examples/cotprompt/README.md create mode 100644 examples/cotprompt/agent/cot_agent/__init__.py create mode 100644 examples/cotprompt/agent/cot_agent/cot_conclude.py create mode 100644 examples/cotprompt/agent/cot_agent/cot_input_interface.py create mode 100644 examples/cotprompt/agent/cot_agent/cot_reasoning.py create mode 100644 examples/cotprompt/compile_container.py create mode 100755 examples/cotprompt/configs/llms/gpt4o.yml create mode 100755 examples/cotprompt/configs/llms/json_res.yml create mode 100755 examples/cotprompt/configs/llms/text_res.yml create mode 100755 examples/cotprompt/configs/tools/all_tools.yml create mode 100755 examples/cotprompt/configs/workers/cot_conclude.yml create mode 100755 examples/cotprompt/configs/workers/cot_input_interface.yml create mode 100644 examples/cotprompt/configs/workers/cot_reasoning.yaml create mode 100644 examples/cotprompt/container.yaml create mode 100644 examples/cotprompt/run_cli.py diff --git a/examples/cotprompt/README.md b/examples/cotprompt/README.md new file mode 100644 index 00000000..47c05897 --- /dev/null +++ b/examples/cotprompt/README.md @@ -0,0 +1,47 @@ +# Chain-of-Thought Prompting Elicits Reasoning in Large Language Models + +## Overview + +This example contains code to implement the basic COT in the paper [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903) by Wei, et al. + +## Running the Example + +For terminal/CLI usage: Input and output are in the terminal window + +```bash +cd examples/cotprompt +python run_cli.py +``` + +An example input and output(by GPT-4o-mini). + +``` +> Input: Which is bigger, 9.9 or 9.11? +> Reasoning output: + +1. **Analyze the Question**: The question asks which of the two numbers, 9.9 or 9.11, is larger. To answer this, we must understand the place values and numerical representation of both numbers. + +2. **Consider the Values**: 9.9 can be broken down into 9 plus 0.9, while 9.11 can be broken down into 9 plus 0.11. This shows that both numbers start with the same whole number part (9). + +3. **Compare the Decimal Parts**: We need to compare the decimal portions: 0.9 and 0.11. The number 0.9 can also be expressed as 0.90 for comparison purposes, thus making it easier to analyze alongside 0.11. + +4. **Evaluate the Decimal Comparison**: Now we compare the decimals: 0.90 (equivalent to 0.9) and 0.11. Here, we observe that 0.90 is greater than 0.11 because in the tenths place, 0.90 has a 9 compared to 0.11’s 1, which indicates a larger value. + +5. **Conclude Comparison Process**: Since both numbers share the same whole number part of 9, the larger value will depend solely on the larger decimal part, which we have determined through comparison. + +At this point, we have thoroughly analyzed the components of each number and identified that the key to determining which is larger lies in the comparison of their decimal parts. + +>Final answer: 9.9 is bigger than 9.11. +``` + +## Citation +``` +@article{wei2022chain, + title={Chain-of-thought prompting elicits reasoning in large language models}, + author={Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten and Xia, Fei and Chi, Ed and Le, Quoc V and Zhou, Denny and others}, + journal={Advances in neural information processing systems}, + volume={35}, + pages={24824--24837}, + year={2022} +} +``` \ No newline at end of file diff --git a/examples/cotprompt/agent/cot_agent/__init__.py b/examples/cotprompt/agent/cot_agent/__init__.py new file mode 100644 index 00000000..e69de29b diff --git a/examples/cotprompt/agent/cot_agent/cot_conclude.py b/examples/cotprompt/agent/cot_agent/cot_conclude.py new file mode 100644 index 00000000..56beb4ef --- /dev/null +++ b/examples/cotprompt/agent/cot_agent/cot_conclude.py @@ -0,0 +1,57 @@ +from omagent_core.engine.worker.base import BaseWorker +from omagent_core.utils.registry import registry +from omagent_core.models.llms.openai_gpt import OpenaiGPTLLM +from omagent_core.models.llms.schemas import Message, Content +from omagent_core.models.llms.base import BaseLLMBackend +from omagent_core.utils.logger import logging +from pathlib import Path + +CURRENT_PATH = Path(__file__).parents[0] + +@registry.register_worker() +class COTConclusion(BaseLLMBackend, BaseWorker): + """根据问题和推理步骤生成最终答案。""" + + llm: OpenaiGPTLLM + + def _run(self, user_question:str, reasoning_result:str, *args, **kwargs): + # 从工作流变量中提取问题和推理结果 + # user_question = kwargs['user_question'] + # reasoning_result = kwargs['reasoning_result'] + + # 生成最终答案的 prompt + final_answer_prompt = self.generate_final_answer_prompt(user_question, reasoning_result) + + # 调用 LLM 生成最终答案 + final_answer = self.call_llm_with_prompt(final_answer_prompt) + + # 将最终答案存储到工作流变量中 + self.stm(self.workflow_instance_id)['final_answer'] = final_answer + + self.callback.send_answer(self.workflow_instance_id, msg=final_answer) + + return {'final_answer': final_answer} + + def generate_final_answer_prompt(self, question: str, reasoning_result: str): + """生成最终答案的 prompt""" + + return f""" + You have completed the reasoning process. Based on the following question and reasoning result, generate a concise and clear final answer. + + Question: {question} + Reasoning steps: {reasoning_result} + + Please provide the final answer: + """ + + def call_llm_with_prompt(self, prompt: str): + """调用 LLM 生成最终答案""" + chat_message = [] + + # Add text question as first message + chat_message.append(Message(role="user", message_type='text', content=prompt)) + + #print("chat_message++++++++++++++++++++++++++++",chat_message) + + response = self.llm.generate(chat_message) + return response['choices'][0]['message']['content'] diff --git a/examples/cotprompt/agent/cot_agent/cot_input_interface.py b/examples/cotprompt/agent/cot_agent/cot_input_interface.py new file mode 100644 index 00000000..c65286ce --- /dev/null +++ b/examples/cotprompt/agent/cot_agent/cot_input_interface.py @@ -0,0 +1,21 @@ +from omagent_core.engine.worker.base import BaseWorker +from omagent_core.utils.registry import registry +from omagent_core.utils.logger import logging +from pathlib import Path + +CURRENT_PATH = Path(__file__).parents[0] + +@registry.register_worker() +class COTInputInterface(BaseWorker): + """处理用户输入问题并将其存储到工作流变量中。""" + + def _run(self, *args, **kwargs): + # 读取用户输入的问题 + user_input = self.input.read_input(workflow_instance_id=self.workflow_instance_id, input_prompt='请输入你的问题:') + + question = user_input['messages'][-1]['content'][0]['data'] + + # 将问题存储到工作流的变量中 + self.stm(self.workflow_instance_id)['user_question'] = question + + return {'user_question': question} diff --git a/examples/cotprompt/agent/cot_agent/cot_reasoning.py b/examples/cotprompt/agent/cot_agent/cot_reasoning.py new file mode 100644 index 00000000..6db2b95b --- /dev/null +++ b/examples/cotprompt/agent/cot_agent/cot_reasoning.py @@ -0,0 +1,64 @@ +from omagent_core.models.llms.openai_gpt import OpenaiGPTLLM +from omagent_core.engine.worker.base import BaseWorker +from omagent_core.utils.registry import registry +from omagent_core.models.llms.base import BaseLLMBackend +from omagent_core.models.llms.schemas import Message, Content +from omagent_core.utils.logger import logging +from pathlib import Path + +CURRENT_PATH = Path(__file__).parents[0] + +@registry.register_worker() +class COTReasoning(BaseLLMBackend, BaseWorker): + """根据问题生成推理过程的中间数据，不给出答案。""" + + llm: OpenaiGPTLLM + + def _run(self, user_question:str,*args, **kwargs): + # 从工作流变量中提取问题 + #user_question = kwargs['user_question'] + + # 生成推理过程的 prompt + reasoning_prompt = self.generate_reasoning_prompt(user_question) + #print("reasoning_prompt++++++++++++++++++++++++++++",reasoning_prompt) + + # 调用 LLM 生成推理步骤 + reasoning_result = self.call_llm_with_prompt(reasoning_prompt) + + # 将推理结果存储到工作流变量中 + self.stm(self.workflow_instance_id)['reasoning_result'] = reasoning_result + + self.callback.send_answer(self.workflow_instance_id, msg=reasoning_result) + + return {'reasoning_result': reasoning_result} + + def generate_reasoning_prompt(self, question: str): + """生成推理过程的 prompt""" + return f""" + Please reason step by step based on the following question and provide the reasoning process: + + question: {question} + + Reasoning Steps: + 1. Analyze the question and consider possible solutions. + 2. Reason through the different aspects of the question step by step. + 3. Only provide the reasoning process, do not give the final answer. + + Please begin reasoning: + """ + + def call_llm_with_prompt(self, prompt: str): + """调用 LLM 生成推理步骤""" + + + chat_message = [] + + # Add text question as first message + chat_message.append(Message(role="user", message_type='text', content=prompt)) + + #print("chat_message++++++++++++++++++++++++++++",chat_message) + + response = self.llm.generate(chat_message) + if response is None: + raise ValueError("LLM inference returned None.") + return response['choices'][0]['message']['content'] diff --git a/examples/cotprompt/compile_container.py b/examples/cotprompt/compile_container.py new file mode 100644 index 00000000..1e8a8659 --- /dev/null +++ b/examples/cotprompt/compile_container.py @@ -0,0 +1,17 @@ +# Import core modules and components +from omagent_core.utils.container import container + +# Import workflow related modules +from pathlib import Path +from omagent_core.utils.registry import registry + +# Set up path and import modules +CURRENT_PATH = root_path = Path(__file__).parents[0] +registry.import_module() + +# Register required components +container.register_callback(callback='AppCallback') +container.register_input(input='AppInput') +container.register_stm("RedisSTM") +# Compile container config +container.compile_config(CURRENT_PATH) diff --git a/examples/cotprompt/configs/llms/gpt4o.yml b/examples/cotprompt/configs/llms/gpt4o.yml new file mode 100755 index 00000000..e7962308 --- /dev/null +++ b/examples/cotprompt/configs/llms/gpt4o.yml @@ -0,0 +1,7 @@ +name: OpenaiGPTLLM +model_id: gpt-4o-mini +api_key: ${env| custom_openai_key, openai_api_key} +endpoint: ${env| custom_openai_endpoint, https://api.openai.com/v1} +temperature: 0 +vision: true +response_format: json_object \ No newline at end of file diff --git a/examples/cotprompt/configs/llms/json_res.yml b/examples/cotprompt/configs/llms/json_res.yml new file mode 100755 index 00000000..8baac86f --- /dev/null +++ b/examples/cotprompt/configs/llms/json_res.yml @@ -0,0 +1,6 @@ +name: OpenaiGPTLLM +model_id: gpt-4o-mini +api_key: ${env| custom_openai_key, openai_api_key} +endpoint: ${env| custom_openai_endpoint, https://api.openai.com/v1} +temperature: 0 +#vision: true \ No newline at end of file diff --git a/examples/cotprompt/configs/llms/text_res.yml b/examples/cotprompt/configs/llms/text_res.yml new file mode 100755 index 00000000..0505fb5c --- /dev/null +++ b/examples/cotprompt/configs/llms/text_res.yml @@ -0,0 +1,6 @@ +name: OpenaiGPTLLM +model_id: gpt-4o-mini +api_key: ${env| custom_openai_key, openai_api_key} +endpoint: ${env| custom_openai_endpoint, https://api.openai.com/v1} +temperature: 0 +response_format: text \ No newline at end of file diff --git a/examples/cotprompt/configs/tools/all_tools.yml b/examples/cotprompt/configs/tools/all_tools.yml new file mode 100755 index 00000000..1352f09e --- /dev/null +++ b/examples/cotprompt/configs/tools/all_tools.yml @@ -0,0 +1,10 @@ +llm: ${sub|text_res} +tools: + - Calculator + - CodeInterpreter + - ReadFileContent + - WriteFileContent + - ShellTool + - name: WebSearch + bing_api_key: ${env|bing_api_key, null} + llm: ${sub|text_res} diff --git a/examples/cotprompt/configs/workers/cot_conclude.yml b/examples/cotprompt/configs/workers/cot_conclude.yml new file mode 100755 index 00000000..1a62cb34 --- /dev/null +++ b/examples/cotprompt/configs/workers/cot_conclude.yml @@ -0,0 +1,5 @@ +name: COTConclusion +llm: ${sub|json_res} +tool_manager: ${sub|all_tools} +output_parser: + name: StrParser diff --git a/examples/cotprompt/configs/workers/cot_input_interface.yml b/examples/cotprompt/configs/workers/cot_input_interface.yml new file mode 100755 index 00000000..34ee61cd --- /dev/null +++ b/examples/cotprompt/configs/workers/cot_input_interface.yml @@ -0,0 +1,5 @@ +name: COTInputInterface +llm: ${sub|json_res} +#tool_manager: ${sub|all_tools} +output_parser: + name: StrParser diff --git a/examples/cotprompt/configs/workers/cot_reasoning.yaml b/examples/cotprompt/configs/workers/cot_reasoning.yaml new file mode 100644 index 00000000..4a1f92a4 --- /dev/null +++ b/examples/cotprompt/configs/workers/cot_reasoning.yaml @@ -0,0 +1,5 @@ +name: COTReasoning +llm: ${sub|json_res} +#tool_manager: ${sub|all_tools} +output_parser: + name: StrParser diff --git a/examples/cotprompt/container.yaml b/examples/cotprompt/container.yaml new file mode 100644 index 00000000..7c35e832 --- /dev/null +++ b/examples/cotprompt/container.yaml @@ -0,0 +1,84 @@ +conductor_config: + name: Configuration + base_url: + value: http://localhost:8080 + description: The Conductor Server API endpoint + env_var: CONDUCTOR_SERVER_URL + auth_key: + value: null + description: The authorization key + env_var: AUTH_KEY + auth_secret: + value: null + description: The authorization secret + env_var: CONDUCTOR_AUTH_SECRET + auth_token_ttl_min: + value: 45 + description: The authorization token refresh interval in minutes. + env_var: AUTH_TOKEN_TTL_MIN + debug: + value: false + description: Debug mode + env_var: DEBUG +connectors: + redis_stream_client: + name: RedisConnector + host: + value: localhost + env_var: HOST + port: + value: 6379 + env_var: PORT + password: + value: null + env_var: PASSWORD + username: + value: null + env_var: USERNAME + db: + value: 0 + env_var: DB + redis_stm_client: + name: RedisConnector + host: + value: localhost + env_var: HOST + port: + value: 6379 + env_var: PORT + password: + value: null + env_var: PASSWORD + username: + value: null + env_var: USERNAME + db: + value: 1 + env_var: DB +components: + RedisSTM: + name: RedisSTM + DefaultCallback: + name: DefaultCallback + bot_id: + value: '' + env_var: BOT_ID + start_time: + value: 2024-11-10_20:51:13 + env_var: START_TIME + folder_name: + value: ./running_logs/2024-11-10_20:51:13 + env_var: FOLDER_NAME + AppInput: + name: AppInput + AppCallback: + name: AppCallback + bot_id: + value: '' + env_var: BOT_ID + start_time: + value: 2024-11-10_20:51:13 + env_var: START_TIME + folder_name: + value: ./running_logs/2024-11-10_20:51:13 + env_var: FOLDER_NAME diff --git a/examples/cotprompt/run_cli.py b/examples/cotprompt/run_cli.py new file mode 100644 index 00000000..76482e54 --- /dev/null +++ b/examples/cotprompt/run_cli.py @@ -0,0 +1,56 @@ +from omagent_core.utils.container import container +from omagent_core.engine.workflow.conductor_workflow import ConductorWorkflow +from omagent_core.engine.workflow.task.simple_task import simple_task +from omagent_core.utils.registry import registry +from omagent_core.clients.devices.cli.client import DefaultClient +from omagent_core.utils.logger import logging +from omagent_core.engine.workflow.task.set_variable_task import SetVariableTask + +from agent.cot_agent.cot_input_interface import COTInputInterface +from agent.cot_agent.cot_reasoning import COTReasoning +from agent.cot_agent.cot_conclude import COTConclusion +from pathlib import Path +logging.init_logger("omagent", "omagent", level="INFO") + +# 设置当前工作目录路径 +CURRENT_PATH = Path(__file__).parents[0] + +# 注册worker +registry.import_module(project_path=CURRENT_PATH.joinpath('agent')) +container.register_stm("RedisSTM") +container.from_config(CURRENT_PATH.joinpath('container.yaml')) + +# 初始化工作流 +workflow = ConductorWorkflow(name='cot_algorithm_workflow') + +# 任务1: 输入接口任务 +client_input_task = simple_task(task_def_name='COTInputInterface', task_reference_name='cot_input_interface') + +# 任务2: 设置 user_question 变量 +# init_set_variable_task = SetVariableTask( +# task_ref_name='init_set_variable_task', +# input_parameters={'user_question': client_input_task.output('user_question')} +# ) + +# 任务3: 推理生成任务 +reasoning_task = simple_task(task_def_name='COTReasoning', task_reference_name='cot_reasoning', inputs={'user_question': client_input_task.output('user_question')}) + +# 任务4: 设置推理结果变量 +# set_reasoning_result_task = SetVariableTask( +# task_ref_name='set_reasoning_result_task', +# input_parameters={'reasoning_result': reasoning_task.output('reasoning_result')} +# ) + +# 任务5: 生成最终答案 +conclude_task = simple_task(task_def_name='COTConclusion', task_reference_name='cot_conclude', inputs={'user_question': client_input_task.output('user_question'), 'reasoning_result': reasoning_task.output('reasoning_result')}) + +# 定义工作流执行顺序: 输入 -> 设置 user_question -> 推理生成 -> 结论生成 +workflow >> client_input_task >> reasoning_task >> conclude_task + +# 注册工作流 +workflow.register(overwrite=True) + +# 启动应用 +config_path = CURRENT_PATH.joinpath('configs') +cli_client = DefaultClient(interactor=workflow, config_path=config_path, workers=[COTInputInterface()]) +cli_client.start_interactor()