Muna client SDKs contain a number of undocumented features, usually targeted towards maximizing performance or improving the developer experience.

​
Monkey Patching in Python

Instead of calling the muna.predictions.create method, the Python SDK provides an @muna.beta.predict decorator which monkey-patches an existing function to instead make a prediction:

​
Creating a Prediction

Apply the @muna.beta.predict decorator to a function. When the function is called, it will instead create a prediction with the provided tag and return the results: 
from muna import Muna

# 💥 Create a Muna client
muna = Muna(...)

# 🔥 Monkey-patch a function
@muna.beta.predict(tag="@yusuf/add")
def add_numbers (a, b):
    ...

# 🚀 Call the function
print(add_numbers(9, 10))

# Output:
# 21

​
Streaming a Prediction

To instead stream the prediction, add an Iterator or Generator return type annotation to the function: 
from muna import Muna

# 💥 Create a Muna client
muna = Muna(...)

# 🔥 Monkey-patch a function
@muna.beta.predict(tag="@yusuf/split-sentence")
def split_sentence (sentence) -> Iterator[str]:
    ...

# 🚀 Call the function
for split in split_sentence("This is a long speech"):
    print(split)

# Output:
# "This"
# "is"
# "a"
# "long"
# "speech"

​
Parsing Pydantic Models

The decorator also supports parsing output objects into Pydantic models by specifying a return type annotation: 
from muna import Muna
from pydantic import BaseModel
from typing import Literal

class Pet (BaseModel):
    sound: Literal["bark", "meow"]
    legs: int

# 💥 Create a Muna client
muna = Muna(...)

# 🔥 Monkey-patch a function
@muna.beta.predict(tag="@yusuf/pet-image-detector")
def detect_pet (image) -> Pet:
    ...

# 🚀 Call the function
print(repr(detect_pet(cat_image)))

# Output:
# Pet(sound="bark", legs=8)