Schema

We use Schema to define the structure of the data. It is a very important concept in Promptulate. The schema is composed of Message and MessageSet classes, which encapsulate the data structure for dialogues and conversations.

Message

A Message is a fundamental building block for dialogues in Promptulate. It represents a single unit of communication in a conversation, whether it's from the system, the user, or the assistant (AI). Each Message object contains the content of the message and any additional metadata that may be relevant. There are four types of messages:

• SystemMessage: Represents messages that provide instructions or context for the conversation, typically not displayed to the end user.
• UserMessage: Represents messages that come from the human user, indicating their input or questions.
• AssistantMessage: Represents messages generated by the AI assistant in response to the user.
• CompletionMessage: Used specifically in the context of OpenAI's API to represent a message that is a completion of a prompt.

You can see detail code in https://github.com/Undertone0809/promptulate/blob/main/promptulate/schema.py#L19C24-L19C24

Each message type inherits from the BaseMessage class, which ensures that every message has a consistent structure and set of properties.

MessageSet

A MessageSet is a collection of Message objects that together represent a full conversation or a segment of a conversation. It is the main data structure used to manage the flow of dialogue between the user and the assistant. MessageSet provides methods for converting the messages to different formats required by various language models (LLMType) or for serialization purposes.

MessageSet includes methods for adding new messages of different types to the conversation, such as add_user_message or add_ai_message, which take the message content as input and create the appropriate Message object.

StreamIterator

A StreamIterator is an iterator for the response stream from the LLM model.StreamIterator provides methods for converting the messages of stream to different formats required by various language models (LLMType) or for serialization purposes.

To use StreamIterator ,three parameters are required.

• response_stream: The stream of responses from the LLM model.
• parse_content: The callback function to parse the chunk.
• return_raw_response: A boolean indicating whether to return the raw response

Example Usage

Here's an example of how you might use MessageSet and Message in a Promptulate application:

import promptulate as pne
from promptulate.schema import AssistantMessage, MessageSet, SystemMessage, UserMessage

messages = MessageSet(
    messages=[
        SystemMessage(content="You are a helpful assitant"),
        UserMessage(content="Hello?"),
    ]
)

llm = pne.ChatOpenAI()
answer: AssistantMessage = llm.predict(messages)
print(answer.content)
print(answer.additional_kwargs)

You can also use MessageSet to convert a OpenAI type messages:

from promptulate.schema import MessageSet

messages: MessageSet = MessageSet.from_listdict_data([
    {"role": "system", "content": "You are a helpful assitant"},
    {"role": "user", "content": "Hello?"},
])
str_message: str = messages.string_messages
openai_type_msg: list = messages.listdict_messages

You can use MessageSet to append messages:

from promptulate.schema import MessageSet, SystemMessage, UserMessage

messages = MessageSet(
    messages=[
        SystemMessage(content="You are a helpful assitant"),
        UserMessage(content="Hello?"),
    ]
)
messages.add_ai_message(content="Hello, how can I help you?")
messages.add_user_message(content="I need help with my computer")

Here's an example of how you might use `BaseStreamIterator` in a `Promptulate` application:

```python
from promptulate.schema import StreamIterator


# we use qianfan sdk as an example
def parse_content(chunk) -> (str, str):
    content = chunk["result"]
    ret_data = chunk["body"]
    return content, ret_data


def stream():
    return_raw_response = True
    # add chunk message in response
    response = ""
    return StreamIterator(
        response_stream=response,
        parse_content=parse_content,
        return_raw_response=return_raw_response,
    )