Building Prompts
Creating Prompts¶
Marvin lets you define dynamic prompts using code, eliminating the need for cumbersome template management. With this approach, you can easily create reusable and modular prompts, streamlining the development process.
Example
from typing import Optional
from marvin.prompts.library import System, User, ChainOfThought
class ExpertSystem(System):
content: str = (
"You are a world-class expert on {{ topic }}. "
"When asked questions about {{ topic }}, you answer correctly."
)
topic: Optional[str]
prompt = (
ExpertSystem(topic="python")
| User("Write a function to find the nth Fibonacci number.")
| ChainOfThought() # Tell the LLM to think step by step
)
# We can now call `dict` to get the formatted messages.
prompt.dict()
Click to see output
[
{
'role': 'system',
'content': 'You are a world-class expert on python.
When asked questions about python, you
answer correctly.'
},
{
'role': 'user',
'content': 'I need to know how to write a function to
find the nth Fibonacci number.'
},
{ 'role': 'assistant',
'content': "Let's think step by step."
}
]
Templating Prompts¶
In many applications, templating is unavoidable. In these cases, Marvin's optional templating engine simplifies the process of sharing context across prompts to an unprecedented level. By passing native Python types or Pydantic objects into the rendering engine, you can seamlessly establish context for entire conversations. This feature enables effortless information flow and context continuity throughout the prompt interactions.
Example
from typing import Optional
from marvin.prompts.library import System, User, ChainOfThought
class ExpertSystem(System):
content: str = (
"You are a world-class expert on {{ topic }}. "
"When asked questions about {{ topic }}, you answer correctly."
)
topic: Optional[str]
prompt = (
ExpertSystem()
| User(
"I need to know how to write a function in {{ topic }} to find the nth Fibonacci "
"number."
)
| ChainOfThought() # Tell the LLM to think step by step
)
# We can now call `dict` with keyword arguments to get the formatted messages.
prompt.dict(topic="rust")
Click to see output
[
{
'role': 'system',
'content': 'You are a world-class expert on rust.
When asked questions about rust, you answer correctly.'
},
{
'role': 'user',
'content': 'I need to know how to write a function in
rust to find the nth Fibonacci number.'
},
{
'role': 'assistant',
'content': "Let's think step by step."
}
]
Example: ReAct¶
Example
from marvin.prompts.library import System
class ReActPattern(System):
content = """
You run in a loop of Thought, Action, PAUSE, Observation.
At the end of the loop you output an Answer
Use Thought to describe your thoughts about the question you have been asked.
Use Action to run one of the actions available to you - then return PAUSE.
Observation will be the result of running those actions.
"""
Example: SQL¶
Example
import pydantic
from marvin.prompts.library import System
class ColumnInfo(pydantic.BaseModel):
name: str
description: str
class SQLTableDescription(System):
content = """
If you chose to, you may query a table whose schema is defined below:
"""
columns: list[ColumnInfo] = pydantic.Field(
..., description="name, description pairs of SQL Schema"
)
UserQueryPrompt = SQLTableDescription(
columns=[
ColumnInfo(name="last_login", description="Date and time of user's last login"),
ColumnInfo(
name="date_created",
description="Date and time when the user record was created",
),
ColumnInfo(
name="date_last_purchase",
description="Date and time of user's last purchase",
),
]
)
print(UserQueryPrompt.read())
Executing Prompts¶
Marvin makes executing one-off task or chain patterns dead simple.
Running a task¶
Once you have a prompt defined, fire it off with your chosen LLM asyncronously like so:
Example
from marvin.prompts.library import System, User, ChainOfThought
from marvin.engine.language_models import chat_llm
from typing import Optional
class ExpertSystem(System):
content: str = (
"You are a world-class expert on {{ topic }}. "
"When asked questions about {{ topic }}, you answer correctly. "
"You only answer questions about {{ topic }}. "
)
topic: Optional[str]
class Tutor(System):
content: str = (
"When you give an answer, you modulate your response based on the "
"inferred knowledge of the user. "
"Your student's name is {{ name }}. "
)
name: str = "not provided"
model = chat_llm()
response = await model(
(
ExpertSystem()
| Tutor()
| User(
"I heard that there are types of geometries when the angles don't add up to"
" 180?"
)
| ChainOfThought()
).render(topic="geometry", name="Adam")
)
print(response.content)
Click to see output
Yes, you are correct! In traditional Euclidean geometry,
the angles of a triangle always add up to 180 degrees.
However, there are indeed other types of geometries where
this is not the case. One such example is non-Euclidean
geometry, which includes hyperbolic and elliptic geometries.
In hyperbolic geometry, the angles of a triangle add up to
less than 180 degrees, while in elliptic geometry, the angles
add up to more than 180 degrees. These non-Euclidean
geometries have their own unique properties and are
studied in mathematics and physics.