Langchain Chat Model 和 Chat Prompt Template

0. 简介

Chat Model 不止是一个用于谈天对话的模子抽象，更紧张的是提供了多角色提示能力（System,AI,Human,Function)。
Chat Prompt Template 则为开发者提供了便捷维护不同角色的提示模板与消息记录的接口。
1. 构造 ChatPromptTemplate

1. from langchain.prompts.chat import (
2.     ChatPromptTemplate,
3.     SystemMessagePromptTemplate,
4.     AIMessagePromptTemplate,
5.    HumanMessagePromptTemplate,
6. )
8. import os
9. from dotenv import load_dotenv, find_dotenv
11. # 删除all_proxy环境变量
12. if 'all_proxy' in os.environ:
13.     del os.environ['all_proxy']
15. # 删除ALL_PROXY环境变量
16. if 'ALL_PROXY' in os.environ:
17.     del os.environ['ALL_PROXY']
19. _ = load_dotenv(find_dotenv())
21. template = (
22.     """You are a translation expert, proficient in various languages. \n
23.     Translates English to Chinese."""
24. )
25. system_message_prompt = SystemMessagePromptTemplate.from_template(template)
26. print(type(system_message_prompt))
27. print(system_message_prompt)
29. human_template = "{text}"
30. human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
31. print(type(human_message_prompt))
32. print(human_message_prompt)
34. print("*"*40)
35. # 使用 System 和 Human 角色的提示模板构造 ChatPromptTemplate
36. chat_prompt_template = ChatPromptTemplate.from_messages(
37.     [system_message_prompt, human_message_prompt]
38. )
39. print(type(chat_prompt_template))
40. print(chat_prompt_template)
42. print("*"*50)
43. chat_prompt_prompt_value = chat_prompt_template.format_prompt(text="I love python.")
44. print(type(chat_prompt_prompt_value))
45. print(chat_prompt_prompt_value)
47. print("*"*60)
48. chat_prompt_list = chat_prompt_template.format_prompt(text="I love python.").to_messages()
49. print(type(chat_prompt_list))
50. print(chat_prompt_list)

复制代码

输出：

1. <class 'langchain_core.prompts.chat.SystemMessagePromptTemplate'>
2. prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese.') additional_kwargs={}
3. <class 'langchain_core.prompts.chat.HumanMessagePromptTemplate'>
4. prompt=PromptTemplate(input_variables=['text'], input_types={}, partial_variables={}, template='{text}') additional_kwargs={}
5. ****************************************
6. <class 'langchain_core.prompts.chat.ChatPromptTemplate'>
7. input_variables=['text'] input_types={} partial_variables={} messages=[SystemMessagePromptTemplate(prompt=PromptTemplate(input_variables=[], input_types={}, partial_variables={}, template='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese.'), additional_kwargs={}), HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['text'], input_types={}, partial_variables={}, template='{text}'), additional_kwargs={})]
8. **************************************************
9. <class 'langchain_core.prompt_values.ChatPromptValue'>
10. messages=[SystemMessage(content='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese.', additional_kwargs={}, response_metadata={}), HumanMessage(content='I love python.', additional_kwargs={}, response_metadata={})]
11. ************************************************************
12. <class 'list'>
13. [SystemMessage(content='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese.', additional_kwargs={}, response_metadata={}), HumanMessage(content='I love python.', additional_kwargs={}, response_metadata={})]

复制代码

2. LCEL 执行

1. from langchain_openai import ChatOpenAI
2. from langchain_core.output_parsers import StrOutputParser
3. from langchain.prompts.chat import (
4.     ChatPromptTemplate,
5.     SystemMessagePromptTemplate,
6.     AIMessagePromptTemplate,
7.    HumanMessagePromptTemplate,
8. )
9. import os
10. from dotenv import load_dotenv, find_dotenv
12. # 删除all_proxy环境变量
13. if 'all_proxy' in os.environ:
14.     del os.environ['all_proxy']
16. # 删除ALL_PROXY环境变量
17. if 'ALL_PROXY' in os.environ:
18.     del os.environ['ALL_PROXY']
20. _ = load_dotenv(find_dotenv())
22. # 为了结果的稳定性，将 temperature 设置为 0
23. translation_model = ChatOpenAI(model_name="gpt-3.5-turbo", temperature=0)
25. template = (
26.     """You are a translation expert, proficient in various languages. \n
27.     Translates {source_language} to {target_language} in the style of {name}."""
28. )
29. system_message_prompt = SystemMessagePromptTemplate.from_template(template)
31. human_template = "{text}"
32. human_message_prompt = HumanMessagePromptTemplate.from_template(human_template)
34. # 使用 System 和 Human 角色的提示模板构造 ChatPromptTemplate
35. m_chat_prompt_template = ChatPromptTemplate.from_messages(
36.     [system_message_prompt, human_message_prompt]
37. )
38. output_parser = StrOutputParser()
40. m_translation_chain = m_chat_prompt_template| translation_model |  StrOutputParser(callbacks=[callback_handler])
42. # Prepare input data
43. input_data = {
44.     "source_language": "English",
45.     "target_language": "Chinese",
46.     "name": "严复",
47.     "text": "Life is full of regrets. All we can do is to minimize them.",
48. }
49. input_data1 = {
50.     "source_language": "English",
51.     "target_language": "Chinese",
52.     "name": "李白",
53.     "text": "Life is full of regrets. All we can do is to minimize them.",
54. }
57. # Format the prompt
58. prompt_value = m_chat_prompt_template.format_prompt(**input_data)
59. print(type(prompt_value))
60. print(prompt_value)
62. print(type(prompt_value.to_messages()))
63. print(prompt_value.to_messages())
65. result = translation_model.invoke(prompt_value)
66. print(result)
67. result = m_translation_chain.invoke(input_data)
68. print(result)
69. result = m_translation_chain.invoke(input_data1)
70. print(result)

复制代码

输出：

1. <class 'langchain_core.prompt_values.ChatPromptValue'>
2. messages=[SystemMessage(content='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese in the style of 严复.', additional_kwargs={}, response_metadata={}), HumanMessage(content='Life is full of regrets. All we can do is to minimize them.', additional_kwargs={}, response_metadata={})]
3. <class 'list'>
4. [SystemMessage(content='You are a translation expert, proficient in various languages. \n\n    Translates English to Chinese in the style of 严复.', additional_kwargs={}, response_metadata={}), HumanMessage(content='Life is full of regrets. All we can do is to minimize them.', additional_kwargs={}, response_metadata={})]
5. content='人生充满了遗憾。我们所能做的就是尽量减少它们。' additional_kwargs={'refusal': None} response_metadata={'token_usage': {'completion_tokens': 31, 'prompt_tokens': 53, 'total_tokens': 84, 'completion_tokens_details': None, 'prompt_tokens_details': None}, 'model_name': 'gpt-3.5-turbo-0125', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None} id='run-676a9818-bbbc-44f7-a30e-0ec065aa502f-0' usage_metadata={'input_tokens': 53, 'output_tokens': 31, 'total_tokens': 84, 'input_token_details': {}, 'output_token_details': {}}
6. 人生充满了遗憾。我们所能做的就是尽量减少它们。
7. 人生充滿遺憾，唯有盡量減少。

复制代码

当 translation_model.invoke(input_data) 输入参数是字典时报错

1. result = translation_model.invoke(input_data)
2. print(result)

复制代码

报错：

1. ValueError: Invalid input type <class 'dict'>. Must be a PromptValue, str, or list of BaseMessages.

复制代码

当 m_translation_chain.invoke(prompt_value) 输入参数是prompt_value时报错

1. result = m_translation_chain.invoke(prompt_value)
2. print(result)

复制代码

报错：

1. TypeError: Expected mapping type as input to ChatPromptTemplate. Received <class 'langchain_core.prompt_values.ChatPromptValue'>.

复制代码

检察属性：

1. input_schema = m_translation_chain.input_schema.model_json_schema()
2. print(input_schema)
3. output_schema = m_translation_chain.output_schema.model_json_schema()
4. print(output_schema)

复制代码

输出：

1. {'properties': {'name': {'title': 'Name', 'type': 'string'}, 'source_language': {'title': 'Source Language', 'type': 'string'}, 'target_language': {'title': 'Target Language', 'type': 'string'}, 'text': {'title': 'Text', 'type': 'string'}}, 'required': ['name', 'source_language', 'target_language', 'text'], 'title': 'PromptInput', 'type': 'object'}
3. {'title': 'StrOutputParserOutput', 'type': 'string'}

复制代码

免责声明：如果侵犯了您的权益，请联系站长，我们会及时删除侵权内容，谢谢合作！更多信息从访问主页：qidao123.com:ToB企服之家，中国第一个企服评测及商务社交产业平台。