Open kaykumar opened 4 years ago
Thanks, this is a good idea. Added to the list of feature requests. Would gladly accept contributions for this.
I can definitely help with the Sagemaker deployment automation piece.
I can definitely help with the Sagemaker deployment automation piece.
That would be great!
Is ludwig already saving the tf model in SavedModel format?
Is ludwig already saving the tf model in SavedModel format?
Yes, Ludwig can export the model as SavedModel, but you need to preprocess the incoming data in the right way and postprocess the model outputs correctly if you want the results to make sense, so uploading only the SavedModel withou pre/post-processing data wouldn't do it.
One option is to use the export Neuropod functionality, with also perfroms preprocessing and postprocessing.
An additional option is to use the ludwig serve
command that spawns a Rest API that performs preprocessing, model inference and postprocessing.
Hi @kaykumar I have created an example on how to build, train and deploy a ludwig model with Amazon SageMaker including building a container from source so you have the latest ludwig code. see: https://github.com/aws-samples/amazon-sagemaker-ludwig-transformer
@brightsparc this is great, thank you so much! Do you mind if we reference this on the website in the exampels section as a deployment example?
@w4nderlust sure would be great to get these referenced.
Is your feature request related to a problem? Please describe. A lot of custom work to deploy the model.
Describe the use case I build this model using ludwig, now It will be great if I can provide my AWS /AZURE credentials and it gets deployed and an endpoint is returned, that will help with quick model deployments.
Describe the solution you'd like Ludwig --experiment -- > Ludwig --deploy aws/azure/gcp
Describe alternatives you've considered Build a docker image and deploy to Kubernetes