Generating content¶

Now that you have your prompt, you can combine it with a model call to generate content. Mirascope provides high-level wrappers around common providers so you can focus on prompt engineering instead of learning the interface for providers. Our high-level wrappers are not required to use our prompts but simply provide convenience if you wish to use it.

Note

This doc uses OpenAI. See supported LLM providers for how to generate content with other model providers like Anthropic, Mistral, Cohere and more.

OpenAICall¶

OpenAICall extends BasePrompt and BaseCall to support interacting with the OpenAI API.

Call¶

You can initialize an OpenAICall instance and call the call method to generate an OpenAICallResponse:

import os

from mirascope import OpenAICall, OpenAICallParams

os.environ["OPENAI_API_KEY"] = "YOUR_OPENAI_API_KEY"


class RecipeRecommender(OpenAICall):
    prompt_template = "Recommend recipes that use {ingredient} as an ingredient"

    ingredient: str

    call_params = OpenAICallParams(model="gpt-3.5-turbo-0125")


response = RecipeRecommender(ingredient="apples").call()
print(response.content)  # prints the string content of the call

The call_params of the OpenAI client is tied to the call (and thereby the prompt). Refer to Engineering better prompts [Add link] for more information.

Async¶

If you are want concurrency, you can use the async function instead.

import asyncio
import os

from mirascope import OpenAICall, OpenAICallParams

os.environ["OPENAI_API_KEY"] = "YOUR_API_KEY"


class RecipeRecommender(OpenAICall):
    prompt_template = "Recommend recipes that use {ingredient} as an ingredient"

    ingredient: str

    call_params = OpenAICallParams(model="gpt-3.5-turbo-0125")


async def recommend_recipes():
    """Asynchronously calls the model using `OpenAICall` to generate a recipe."""
    return await RecipeRecommender(ingredient="apples").call_async()


print(asyncio.run(recommend_recipes()))

CallResponse¶

The call method returns an OpenAICallResponse class instance, which is a simple wrapper around the ChatCompletion class in openai that extends BaseCallResponse. In fact, you can access everything from the original response as desired. The primary purpose of the class is to provide convenience.

from mirascope.openai.types import OpenAICallResponse

response = OpenAICallResponse(...)

completion.content     # original.choices[0].message.content
completion.tool_calls  # original.choices[0].message.tool_calls
completion.message     # original.choices[0].message
completion.choice      # original.choices[0]
completion.choices     # original.choices
response.response      # ChatCompletion(...)