Training deep learning models on AWS and GCP instances
AWS provider changes:
spotty sh
command when an instance doesn't have a public one (#112)dockerDataRoot
parameter fixspotInstance: true
parameter for AWS and preemptibleInstance: true
for GCP.spotty run
command:
-S, --sync
flag was replaced with the --no-sync
flag).spotty exec
command to execute custom commands in the container (for example, to run python scripts in the container with PyCharm).-C
flag to the spotty start
command to start or restart a container without restarting the instance itself.instanceProfileArn
parameter to specify custom instance profiles for AWS instances (@tsdalton, #42).spotty.yaml
file (#44).cfn-init
logs automatically downloaded to the local machine if the instance failed to start (#52, #44, #48).spotty.override.yaml
file. It overrides the values of the main spotty.yaml
file and supposed to be added to the .gitignore
file.spotty ssh
command was renamed to spotty sh
as it didn't make sense for the "local" provider and also it's shorter.ports
parameter was moved from the container config to the instance config.ports
parameter to the container configuration to publish specific ports to the host OS).runAsHostUser
parameter to run containers as a host user.-u
flag to spotty run
and spotty sh
commands to connect to the container as a root user.GCP:
bootDiskSize
parameterAWS:
Bug fixes: