FlyMyAI CLI

CLI app provides functionality to deploy custom models onto fma infrastructure. It employs features of pydantic library extended with custom field types to describe model's io. It interacts with fma services to send your models code and to run it on fma's hardware.

Quickstart

Initialization

You have to create a new folder and run fma init <your_model_name> inside of it. This will create one more folder with your model name and a model.py file inside of it. The model.py will be prepolutated with:

from typing import List

from pydantic import BaseModel

from fma.toolkit import model as fma_model


class Input(BaseModel):
    pass


class Output(BaseModel):
    pass


class Model(fma_model.Model):
    requirements: List[str] = []

    def initialize(self):
        pass

    def predict(self, input: Input) -> Output:
        return {}

IMPORTANT NOTE: even though io is described via pydantic models they are not actually supported in the predict method. Instead predict method accepts and returns python dicts. Support for pydantic models will be added eventually.

Model development

During this step you simply have to populate the initialize and predict methods with inferences logic. You can set instance attributes via accessing self, all the imports should be inside of the methods. You can also specify required libraries and packages in the requirements field of Model class.

Deploying the model

If you have finished all the steps above all you need to do is to:

Log in

You can do this via fma login --username <username> --password <password> or by providing one/none, missing ones will be prompted.

Actually deploy

By running fma deploy After this the data about your model will be written into metadata.yaml file.

Extra info

Input fields default values

Fields in input object also support default value declaration. Example of Input class that uses default values:

class Input(BaseModel):
    text: str = Field(default="Hello")
    number: int = Field(default=1)
    fraction: float
    flag: bool = Field(default=True)
    image: Image
    file: File
    audio: Audio
    otext: Optional[str]
    onumber: Optional[int] = Field(default=1)
    ofraction: Optional[float] = Field(default=0.2)
    oflag: Optional[bool] = Field(default=False)
    oimage: Optional[Image]
    ofile: Optional[File]
    oaudio: Optional[Audio]

Additional fields

Currenly standard python str, int, bool, float are available.
The also can be used in combination with typing.Optional
You can specify audio, image and file fields in the io classes by:

from fma.toolkit.fields.audio import Audio
from fma.toolkit.fields.image import Image
from fma.toolkit.fields.file import File

Note that File is currently supported only as the output

Deleting deployed model

You can delete the deployed model if you don't need it anymore: fma delete

Updating the model

fma update will delete the model and redeploy it under the same name if available.

Fetching startup logs

fma logs --output-file logs.tx will download logs into specified file or display them in stdout otherwise.