Prompt Capture¶
Expression statements within APPL functions are captured as prompts based on the type of the value, when the value is not None.
The types can be categorized into three groups according to the behavior when converting to prompts: prompts, sequence of prompts, and custom promptable.
1. Prompts¶
The most basic prompt is a string (including StringFuture), which can be a string literal, formatted string or string variable. We convert the string to a message based on the role scope.
An already constructed Message object is also considered a prompt, as shown in the following example.
import appl
from appl import ppl, records, AIMessage
appl.init()
@ppl
def greeting(name: str):
"Hello world!"
AIMessage("Nice to meet you.")
return records()
print(greeting("APPL"))
# user: Hello world!
# assistant: Nice to meet you.
Besides, for LMs that support multimodal inputs, the Image are also captured as part of the prompt.
Note types like int, float and bool are not captured. To capture these values, you can explicitly convert them to strings or use them in formatted strings.
2. Sequence of Prompts¶
For types that subclass Sequence, such as list and tuple, the elements are recursively captured as prompts one by one.
import appl
from appl import ppl, records
appl.init()
@ppl
def greeting(name: str):
["Hello world!", ("My name is APPL.", "Nice to meet you.")]
return records()
print(greeting("APPL"))
# Hello world!
# My name is APPL.
# Nice to meet you.
3. Custom Promptable¶
You may also define custom types and the ways to convert them to prompts by subclassing Promptable and implementing the __prompt__ method.
import appl
from appl import Promptable, ppl, records
appl.init()
class MyPromptable(Promptable):
def __init__(self, number: int):
self.number = number
def __prompt__(self):
return f"The number is {self.number}."
@ppl
def custom_promptable():
MyPromptable(42)
return records()
print(custom_promptable())
# The number is 42.