如何在LangChain的my_chain中动态访问并修改examples？

假设我们已通过如下代码，基于my_schema创建了LangChain链my_chain：

from langchain.chat_models import ChatOpenAI
from kor.extraction import create_extraction_chain
from kor.nodes import Object, Text, Number
from langchain.chat.models import ChatOpenAI
from langchain.llms import OpenAI

schema = Object(
    id="bank_statement_info",
    description="bank statement information about a given person.",
    attributes=[
        Text(
            id="first_name",
            description="The first name of the person",
            examples=[("John Smith", "John")],
        ),
        Text(
            id="last_name",
            description="The last name of the person",
            examples=[("John Smith", "Smith")],
        ),
        Text(
            id="account_number",
            description="Account Number of the person in Bank statement.",
            examples=[("Account Number: 122-233-566-800", "122-233-566-800")],
        ),
        Text(
            id="address",
            description="address of the person in Bank statement.",
        ),
        Text(
            id="opening_balance",
            description="opening blance of the person in Bank statement.",
            examples=[("opening Balance: 245,800.00","245,800.00")]
        ),
        Text(
            id="closing_balance",
            description="closing blance of the person in Bank statement.",
            examples=[("Closing Balance: 591,800.00","591,800.00")]
        ),
    ],
    examples=[
        (
            """ya 231 Valley Farms Street
              FIRST Santa Monica, CA 90403 STATEMENT OF ACCOUNT
              CITIZENS __firstcitizensbank@domain.com
              BANK
              Account Number: 122-233-566-800
              Statement Date: 11/22/2019 Page 1 of 1
              Period Covered: 05/22/2019 to 11/22/2019
              Eric Nam Opening Balance: 175,800.00
              240 st, apt 15, hill road, Total Credit Amount: 510,000.00
              Baverly Hills, LA, 90209 Total Debit Amount: 94,000.00
              Closing Balance: 591,800.00
              Branch - Baverly Hills Account Type: Saving Account""",
            [
                {"first_name": "John", "last_name": "Smith", "account_number": '122-233-566-800',"address":"240 st, apt 15, hill road,Baverly Hills, LA, 90209",
                 "Closing Balance": "458,589.00","opening Balance": "800.00"},
                {"first_name": "Jane", "last_name": "Doe", "age": '923-533-256-205',"address":"850 st, apt 82, hill road,Baverly Hills, New york, 82044",
                 "Closing Balance": "1000.00","opening Balance": "125,987.00"},
            ],
        )
    ],
    many=True,
)
llm = ChatOpenAI(temperature=0.0, openai_api_key='', open_api_base='', model_kwargs={'engine': 'openai_gpt_4'}
my_chain = create_extraction_chain(llm, my_schema, encoder_or_encoder_class='json')

我希望在创建my_chain后访问其examples，尝试过访问my_chain.prompt但无法找到对应的examples，尤其想实现动态修改my_chain的examples，请问该需求是否可行？

可行，具体操作方式如下：

1. 访问链中的examples #

创建后的my_chain会保留对原始schema的引用，你可以直接通过my_chain.schema访问整个schema结构，进而获取各级别的examples：

# 获取Object节点级别的示例
object_level_examples = my_chain.schema.examples

# 获取单个属性的示例（比如first_name的示例）
first_name_examples = my_chain.schema.attributes[0].examples

# 获取opening_balance的示例
opening_balance_examples = my_chain.schema.attributes[4].examples

2. 动态修改examples #

直接修改my_chain.schema对应的examples属性即可，修改后的示例会在后续调用链时生效：

# 修改Object级别的示例
new_object_example = (
    """新的银行账单文本：
    Account Number: 999-888-777-666
    Lily Evans Opening Balance: 200,000.00
    123 Main St, New York, NY 10001
    Closing Balance: 250,000.00""",
    [
        {"first_name": "Lily", "last_name": "Evans", "account_number": "999-888-777-666",
         "address": "123 Main St, New York, NY 10001", "opening_balance": "200,000.00", "closing_balance": "250,000.00"}
    ]
)
my_chain.schema.examples = [new_object_example]

# 修改单个属性的示例（比如更新closing_balance的示例）
my_chain.schema.attributes[5].examples = [("Closing Balance: 300,000.00", "300,000.00")]

注意事项 #

• 由于create_extraction_chain会直接引用传入的schema对象，修改my_chain.schema会同时改变原始的schema对象。如果需要保留原始schema的完整性，建议先深拷贝一份再修改，之后可以基于拷贝后的schema重新创建链：

import copy

# 深拷贝原始schema
copied_schema = copy.deepcopy(my_chain.schema)
# 修改拷贝后的schema的示例
copied_schema.examples = [new_object_example]
# 基于新schema创建链
new_chain = create_extraction_chain(llm, copied_schema, encoder_or_encoder_class='json')

内容的提问来源于stack exchange，提问作者singa1994