Training deep learning models on AWS and GCP instances.
Spotty drastically simplifies training of deep learning models on AWS and GCP:
- it makes training on GPU instances as simple as training on your local machine
- it automatically manages all necessary cloud resources including images, volumes, snapshots and SSH keys
- it makes your model trainable in the cloud by everyone with a couple of commands
- it uses tmux to easily detach remote processes from their terminals
- it saves you up to 70% of the costs by using AWS Spot Instances and GCP Preemtible VMs
Spotty is an abstraction over any service that can run containers. It can be AWS, GCP, Azure or YC instances, it can be services like AWS Batch or AWS Sagemaker, it can be your local machine or any remote machine with the Docker installed.
At my best knowledge, there are no projects like this.
At the moment, Spotty supports only AWS and GCP providers, but with the next big release this month Spotty will also start supporting "local" and "remote" providers to work with Docker containers locally and through SSH on any remote instance.
The project welcomes any contributors. There are many features that can be added, for example:
- Other cloud providers: Azure, Yandex.Cloud, Alibaba Cloud
- Cloud services: AWS Batch, AWS Sagemaker
- X11 forwarding
- Windows support
If you're interested in contributing, you can create an issue on GitHub or contact me directly in Slack (@opolosin).