amitjslearn / ops-ml

MLOps

sample repo for MLOps

file	description
train_mnist.ipynb	trained using fastai
inference.py	python inference
app.py	deploy using fastapi
export.pkl	fastai model
req_mnist.txt	requirements (not all see Dockerfile)
screenshots/curl_resp_mlops.png	screenshot of output response using curl
screenshots/fastapi_post.png	screenshot of output response using fastapi (/docs) equivalent to postman
Dockerfile	use for building docker image
docker-compose.yaml	alternate for docker build
3.jpg	sample file for testing

build docker image

sudo docker build -t mnist:latest .

launch the container

sudo docker run -p 8000:8000 --name mnist mnist:latest

OR

sudo docker-compose up

Results

using fastapi/postman equivalent

using curl

Criterias

Criterias met	reasons
1. Low latency for inference	the code is structured in oop format so as to make the inference fast, also the model used in mobilenet v2 for transfer learning which is having less number of parameters which supports faster inference with better results/accuracy
2. Low inference docker image size	docker is built on top of python 3.8 image, also fastai is installed without any deps
3. Code structure	the code is in Object oriented format for better abstraction and code is seperated and isolated for particular tasks in a different files