Faster model training and experimentation with Vertex AI | C2C Community

Faster model training and experimentation with Vertex AI


Userlevel 5
Badge +4

When developing a machine learning model, running multiple Trials of your training code with different values can be very time-consuming. Serverless hyperparameter tuning and multi-GPU setup are two services to make model training faster with Vertex AI, Google’s managed AI platform.

 

Nikita Namjoshi demonstrates in this I/O 2022 video how to use these features. They are real time-savers.

https://io.google/2022/program/934cbc5f-42bb-4f6c-99fe-4972995eb381/


3 replies

Userlevel 7
Badge +65

That’s true @MarkusK.

The code tuning to get better results is a pain. Or was a pain. Now Vertex AI can help us with that. But if you know how to code and you have team that can write great TensorFlow or PyTorch code, would you use it? 🤔

Userlevel 5
Badge +4

@ilias I think you need to run your models with different parameters to improve results no matter how good your code is. In the video the accuracy improved from 0.5 to 0.82 with just 6 trials. 😀

Userlevel 7
Badge +65

WOW

Thanks, @MarkusK. I still haven’t found the time to “play” with Vertex AI. But I’m planning to do.

Reply