Custom Prompt

Here, we’ll see how to customize the instruction segment of the Kor prompt.

from kor.extraction import create_extraction_chain
from kor.nodes import Object, Text, Number
from langchain.chat_models import ChatOpenAI
from langchain.prompts import PromptTemplate

llm = ChatOpenAI(
    model_name="gpt-3.5-turbo",
    temperature=0,
)

schema = Object(
    id="person",
    description="Personal information",
    examples=[
        ("Alice and Bob are friends", [{"first_name": "Alice"}, {"first_name": "Bob"}])
    ],
    attributes=[
        Text(
            id="first_name",
            description="The first name of a person.",
        )
    ],
    many=True,
)

Create a template

Here we create an instruction template.

The template accepts 2 optional parameters:

1. type_description – will be replaced with the schema type-descriptor.

2. format_instructions – will be replaced with the format instructions of whichever encoder is used.

instruction_template = PromptTemplate(
    input_variables=["format_instructions", "type_description"],
    template=(
        "[Pep talk for your LLM goes here]\n\n"
        "Add some type description\n\n"
        "{type_description}\n\n"  # Can comment out
        "Add some format instructions\n\n"
        "{format_instructions}\n"
        "Suffix heren\n"
    ),
)


chain = create_extraction_chain(llm, schema, instruction_template=instruction_template)

print(chain.prompt.format_prompt(text="hello").to_string())

[Pep talk for your LLM goes here]

Add some type description

```TypeScript

person: Array<{ // Personal information
 first_name: string // The first name of a person.
}>
```


Add some format instructions

Please output the extracted information in CSV format in Excel dialect. Please use a | as the delimiter. 
 Do NOT add any clarifying information. Output MUST follow the schema above. Do NOT add any additional columns that do not appear in the schema.
Suffix heren


Input: Alice and Bob are friends
Output: first_name
Alice
Bob

Input: hello
Output:

Custom Encoder and TypeDescriptor

from kor import JSONEncoder, TypeDescriptor

class CatEncoder(JSONEncoder):
    def get_instruction_segment(self) -> str:
        return "Encode your response as Cat JSON enclosed in <😼> tags."

class CatType(TypeDescriptor):
    def describe(self, node: Object) -> str:
        """Describe the schema of the node."""
        return f"A 😼 ate the schema of {type(node)} 😼"

instruction_template = PromptTemplate(
    input_variables=["format_instructions", "type_description"],
    template=(
        "[Pep talk for your LLM goes here]\n\n"
        "Add some type description\n\n"
        "{type_description}\n\n"  # Can comment out
        "Add some format instructions\n\n"
        "{format_instructions}\n"
        "Suffix heren\n"
    ),
)


chain = create_extraction_chain(
    llm,
    schema,
    instruction_template=instruction_template,
    encoder_or_encoder_class=CatEncoder,
    type_descriptor=CatType(),
)

print(chain.prompt.format_prompt(text="hello").to_string())

[Pep talk for your LLM goes here]

Add some type description

A 😼 ate the schema of <class 'kor.nodes.Object'> 😼

Add some format instructions

Encode your response as Cat JSON enclosed in <😼> tags.
Suffix heren


Input: Alice and Bob are friends
Output: <json>{"person": [{"first_name": "Alice"}, {"first_name": "Bob"}]}</json>
Input: hello
Output: