ModelKit lets users generate predictions on pre-trained machine learning models in the cloud.

💡 Sorta like npm, but for ML models.

It follows a pretty straightforward approach to run models for now, but there are plans to support image classification and other deep learning models too in the later versions.

ModelKit Server 🔧

This is the server for ModelKit. It's built on FastAPI. It handles JWT auth, simple feed, disk storage, running predictions and model creation utilities for the client.

Endpoints

/feed

GET

Summary

Feed

Parameters

Name	Located in	Description	Required	Schema
start	query		No
count	query		No

Responses

Code	Description
200	Successful Response
422	Validation Error

/predict

POST

Summary

Train

Responses

Code	Description
200	Successful Response
422	Validation Error

/login

POST

Summary

Login

Responses

Code	Description
200	Successful Response
422	Validation Error

/signup

POST

Summary

Signup

Responses

Code	Description
200	Successful Response
422	Validation Error

/model/create

POST

Summary

Create Model

Responses

Code	Description
200	Successful Response

/model/fetch-all

GET

Summary

Fetch All

Parameters

Name	Located in	Description	Required	Schema
username	query		Yes	string

Responses

Code	Description
200	Successful Response
422	Validation Error

/model/fetch-one

GET

Summary

Fetch One

Parameters

Name	Located in	Description	Required	Schema
uid	query		Yes	string

Responses

Code	Description
200	Successful Response
422	Validation Error

/model/upload

POST

Summary

Create Model

Responses

Code	Description
200	Successful Response
422	Validation Error

/

GET

Summary

Home

Responses

Code	Description
200	Successful Response

Models

Body_create_model_model_upload_post

Name	Type	Description	Required
model	binary		Yes

HTTPValidationError

Name	Type	Description	Required
detail	[ object ]		No

ValidationError

Name	Type	Description	Required
loc	[ string ]		Yes
msg	string		Yes
type	string		Yes

app__routes__auth__login__RequestBody

Name	Type	Description	Required
username	string		Yes
password	string		Yes

app__routes__auth__signup__RequestBody

Name	Type	Description	Required
username	string		Yes
name	string		Yes
password	string		Yes

app__routes__predict__RequestBody

Name	Type	Description	Required
model_uid	string		Yes
xtest	[ ]		Yes

Todos

• DRY approach to add JWT validation middleware to specific routes (currently using WET).
• Safely extract user-provided binaries without running code on the server.
• Use a dedicated storage server for saving model binaries.
• Edit query for profiles and model details.

Contributing

Please open an issue if you see a mistake or something that can be improved.