Create Bento Archive

This guide shows you how to create a Bento Archive and export it (bentoml export) to an S3 bucket.

Before You Start #

• Ensure you have completed Developer System Setup.
  • You may need to install bentoml[aws] if you have not done so already.
• This guide assumes you have already trained a model and are ready to create a Bento.

How to Create a Bento Archive #

Create Bento #

Prepare your model and save it as a Bento:

bentoml build -o tag

Create S3 Bucket #

1. Create an AWS profile:

   aws --profile my-aws-profile configure

   Copy

   AWS Access Key ID [None]: AKIAYLGG12ZIQZLL6PUG
   AWS Secret Access Key [None]: AeazKuGrL4UgiUABVnfhwW02oNtTeB4ec9cQwE1y
   Default region name [None]:
   Default output format [None]:

   Copy
2. Create an S3 bucket:

   aws --profile aioli-demo s3 mb s3://mlis-bucket --region us-east-1 
   make_bucket: USER-bucket

   Copy
3. List your bentos:

   bentoml list

   Copy

    Tag                                  Size       Model Size  Creation Time       
   pytorch_mnist_demo:fml2z4xekol6llg6  20.63 KiB  1.30 MiB    2024-03-17 07:40:49 

   Copy

Export Bento as Archive #

1. Export the Bento as an archive using the bentoml export command. The Bento archive must end with .bento:

   bentoml export pytorch_mnist_demo:latest s3://mlis-bucket/bentoml/pytorch_mnist_demo_.bento

   Copy
   Tip

   If this command fails to export the Bento to S3 storage, make sure you have installed bentoml with aws support (e.g., pip install 'bentoml[aws]').
2. Verify the Bento archive is accessible:

   aws --profile aioli-demo s3 ls s3://mlis-bucket/bentoml/

   Copy

   2024-03-18 14:46:19    1255948 pytorch_mnist_demo.bento

   Copy