[Question]: Use Prompt Template with Chat Engine

[anantgupta129]

Question Validation

• [X] I have searched both the documentation and discord for an answer.

Question

Will ReAct Agent is noticed that using prompt template was mor reliable and controlled output compared to system prompt

agent = ReActAgent.from_tools(tool_spec, llm=llm, verbose=True)
agent.update_prompts({"agent_worker:system_prompt": REACT_SYSTEM_PROMPT})
agent.reset()

How to use prompt template with index as chat engine or simple chat engine

index = VectorStoreIndex.from_vector_store(vector_store=vector_store)
engine = index.as_chat_engine(
    chat_mode="condense_plus_context", streaming=True, system_prompt=system_prompt
)

chatEngine = SimpleChatEngine.from_defaults(streaming=True) instead i want to use prompt template with chat engine

[dosubot[bot]]

To use a prompt template with the chat engine instead of the system prompt, you can define a custom chat prompt template using ChatPromptTemplate and then format it for use with the chat API. Here is an example:

from llama_index.legacy.prompts import ChatPromptTemplate, ChatMessage, MessageRole

# Define the message templates
message_templates = [
    ChatMessage(content="You are an expert system.", role=MessageRole.SYSTEM),
    ChatMessage(
        content="Generate a short story about {topic}",
        role=MessageRole.USER,
    ),
]

# Create the chat prompt template
chat_template = ChatPromptTemplate(message_templates=message_templates)

# Format the messages for the chat API
messages = chat_template.format_messages(topic="a brave knight")

# Alternatively, convert to a text prompt for the completion API
prompt = chat_template.format(topic="a brave knight")

Alternatively, you can use the ChatPromptTemplate.from_messages method for a more concise setup:

from llama_index.legacy.prompts import ChatPromptTemplate

# Define the message templates using from_messages
message_templates = [
    ("system", "You are an expert system."),
    ("user", "Generate a short story about {topic}"),
]

# Create the chat prompt template
chat_template = ChatPromptTemplate.from_messages(message_templates)

# Format the messages for the chat API
messages = chat_template.format_messages(topic="a brave knight")

# Alternatively, convert to a text prompt for the completion API
prompt = chat_template.format(topic="a brave knight")

This code snippet demonstrates how to create a custom chat prompt template and format it for use with the chat API. The ChatPromptTemplate class allows you to define a sequence of ChatMessage objects with specified roles and content, which can then be formatted with specific variables for use in the chat engine. The from_messages method provides a more concise way to define the chat prompt template.

Additionally, you can use advanced features such as partial formatting and output parsers. For example:

from llama_index.legacy.prompts import ChatPromptTemplate, ChatMessage, MessageRole
from llama_index.legacy.types import BaseOutputParser

class MockOutputParser(BaseOutputParser):
    def __init__(self, format_string: str) -> None:
        self._format_string = format_string

    def parse(self, output: str) -> Any:
        return {"output": output}

    def format(self, query: str) -> str:
        return query + "\n" + self._format_string

output_parser = MockOutputParser(format_string="output_instruction")

chat_template = ChatPromptTemplate(
    message_templates=[
        ChatMessage(
            content="This is a system message with a {sys_param}",
            role=MessageRole.SYSTEM,
        ),
        ChatMessage(content="hello {text} {foo}", role=MessageRole.USER),
    ],
    output_parser=output_parser,
)

messages = chat_template.format_messages(
    text="world", foo="bar", sys_param="sys_arg"
)

This example shows how to integrate an output parser with your chat prompt template, allowing for more complex formatting and parsing of the chat messages ^{[1][2][3][4][5]}.

To continue talking to Dosu, mention @dosu.