When engaging in ML research, we all want to concentrate on what’s important — designing and implementing our models. But, in reality, we’re forced to spend far too much time on other incidental tasks, like setting servers, manually allocating GPUs, and tracking each model’s ML training.
Now, however, there’s good news: NSML can take care of all those tedious details, freeing you up to focus fully on your models and data.
- - Throw your data into a GPU cloud via a simple command,
- - Run your codes using deep-learning libraries like Tensorflow, PyTorch, and MXNet on a GPU cloud via another simple command, and
- - Get the results as soon as they are done, releasing resources (especially precious GPUs) for other tasks.
And that’s just a start. NSML can accomplish much more. For example, you can compare many different hyperparameter models and even change a model’s hyperparameters while training is ongoing. Your models can also be paused at any time and resumed after modifying the hyperparameters or other settings.
The results of your modeling experiments are automatically recorded and ordered into NSML’s Kaggle-like leaderboard, allowing you to immediately compare your model’s performance with others. NSML also provides instant serving for your models automatically via RESTful APIs.
In short, NSML is an essential system for all ML developers/researchers who want to focus 100% of their resource on developing ML/DL models.
If this sounds like a system you could use, we invite you to apply to our NSML alpha testing program. Just submit your email address and a brief introduction at the link below, and once NSML is ready, you could get an opportunity to be an alpha-tester.
(Please note that, due to the large number of applications, not everyone will be able to participate).