yamlflow 0.0.6

Yet Another ML flow

yamlflow

Yet Another ML flow

STATUS NOT READY

We follow convention over configuration (also known as coding by convention) software design paradigm.

Here are some of the features the yamlflow provides.

1. Build and publish your ML solution as a RESTful Web Service with yaml.

   • You don't need to write web realated code, or dockerfiles.

   • You don't need to benchmark which python web server or framework is best in terms of performance.

   • WE do it for you. All the best, packed in.

Project structure

examle-project
    ...
    ...
    yamlflow.yaml
    predictor.py
    requirements.txt

example yamlflow.yaml

kind: Service  # manifest type, `Retrainig` will be added soon
meta:
  name: ml-project # name of your project
  version: 0.1.0   # version of your project
backend:
  runtime: torch # options are torch, openvino, tensorflow, tensorrt
  device: cpu    # options are cpu, gpu
frontend:
  predictor: predictor.py # path to predictor.py file
  requirements: requirements.txt # path to requirements.txt file

example predictor.py

import os
import torch
from torchvision import models


class Predictor:
    """
    """
    def __init__(self):
        """ Model object initialization.
        """
        self.model = models.resnet18(pretrained=True)


    def pre_process(self, request: dict) -> torch.Tensor:
        """ Pre_process request given from HTTP call content/type,
        best performance: python_numpy_numba.
        """
        shape = request["data"]
        return torch.randn(shape)

    
    def predict(self, model_input: torch.Tensor) -> torch.Tensor:
        """Can run native in python or using 
        inference servers where no python dependency exists.
        """
        with torch.no_grad():
            return self.model(model_input)
    

    def post_process(self, model_output: torch.Tensor) -> dict:
        """Post_process model_output given torch.
        """
        return {"data": [model_output.cpu().detach().tolist()]}

User guide

pip install yamlflow
yamlflow init
yamlflow build -f flow.yaml

Developer guide

pyenv install 3.8.6
poetry env use ~/.pyenv/versions/3.8.6/bin/python
poetry shell
poetry install