1
$\begingroup$

Serverless technology can be used to deploy ML models to production, since the deployment package sizes can be compressed if too large (or built from source with unneeded dependencies stripped).

But there is also the use case of deploying ML for training, not just inference. For example, if a company wanted to allow power users to retrain a model from the front-end.

Is this feasible for Lambda given the long training times?

Whereas latency wouldn't be issue (cold start delay is fine) the runtime could be fairly long (hours).

$\endgroup$
1
  • 1
    $\begingroup$ You can create a lambda function to spin off a EC2 instance which will have long training hours, I dont think there will be any problem. Since I know people used lambda functions for training and cross-validation etc $\endgroup$ Feb 3 at 8:03
1
$\begingroup$

I've used ECS (Fargate) to train models, the retraining trigger could be the start of an ECS service. While ECS has a little latency, it handles well long runtimes.

You can then serve the model via a lambda.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.