AutoML with PyTorch and Ray - Aniket Maurya, Grid.ai

Gradsflow is an open source AutoML library based on PyTorch. It provides automatic model building and training for various deep learning tasks like image classification and text classification. Furthermore, it leverages Ray for hyperparameter tuning and scaling training on a laptop or a cluster of machines. In this talk, we will look at the internals of Gradsflow with Ray and how to tune hyperparameters of your model quickly.

Aniket is a machine learning software engineer and is currently a developer advocate at Grid.ai.

Transcript

1. Democratizing AI
   Low-Code AutoML Library
   Aniket Maurya / @aniketmaurya
   Gradsflow
   

   View Slide

2. About me
   Aniket Maurya
   An open-source AutoML & PyTorch Model
   Training Library
   Developer Advocate
   at
   

   View Slide

3. Vision of Gradsflow
   ● Anyone can Build with AI
   ● Low Code
   ● Save development time
   ● Focus on Solving Problems
   ● Automate Repetitive Tasks
   

   View Slide

4. Who is it for?
   ● Developers
   ● Novice to Expert ML Engineers
   

   View Slide

5. About GradsFlow
   ● Powerful & easy-to-extend Model Training
   API
   ● Automatic Machine Learning model building
   & training
   ● Automatic Hyperparameter tuning & model
   selection.
   ● Increased Productivity by removing
   repetitive tasks.
   

   View Slide

6. About GradsFlow
   ● Powerful & easy-to-extend Model Training
   API
   ● Automatic Machine Learning model
   building & training
   ● Automatic Hyperparameter tuning & model
   selection.
   ● Increased Productivity by removing
   repetitive tasks.
   

   View Slide

7. About GradsFlow
   ● Powerful & easy-to-extend Model Training
   API
   ● Automatic Machine Learning model building
   & training
   ● Automatic Hyperparameter tuning &
   model selection.
   ● Increased Productivity by removing
   repetitive tasks.
   

   View Slide

8. About GradsFlow
   ● Powerful & easy-to-extend Model Training
   API
   ● Automatic Machine Learning model building
   & training
   ● Automatic Hyperparameter tuning & model
   selection.
   ● Increased Productivity by removing
   repetitive tasks.
   

   View Slide

9. Core
   ● Model
   ● Tracker
   ● Tuner
   ● AutoClassifier
   ● AutoSummarization
   ● AutoTask
   High Level API
   Model Training Auto ML
   BaseModel Metrics
   ● AutoDataset
   ● RayDataset
   Data
   Gradsflow
   

   View Slide

10. Automatic Model Building and Training
    

    View Slide

11. Simplified Hyperparameter tuning API
    

    View Slide

12. View Slide

13. Contribute to GradsFlow
    ● New Idea
    ● Improve Documentation
    ● Add Examples
    ● Fix issues
    ● Find/Fix Bugs
    ● Add new feature
    ● Write test cases
    

    View Slide