Thank you for listening :)

HuggingFace is amazing and I totally forgot to mention it as an alternative. If you push your model to HuggingFace and make it public, then you can use their Inference API with Intel Xeon CPU for free. If you want to keep it private or use GPUs, then you will need to use their paid Inference Endpoints. And for both cases HuggingFace offers a really nice Python client. For AWS Lambda you can get maximum 6 CPU cores which is comparable with the lower end Intel Xeon.

IaC tools are tools like Terraform, or CloudFormation or CDK. Most of them have prerequisites that make them impossible to use without paying the vendor behind them. Does that answer the question?

@@DagsHub no, does AWS make Lambda a trap because of S3?

@@staticalmo That's one way to look at it. It's more like, they structure it so that you can't easily use Lambda for free even though they have a free tier, because you need to use S3

​@@staticalmo When AWS Serverless Application Model (SAM) tool deploys Lambda infrastructure code, it first creates a CloudFormation changeset and always stores it in an S3 bucket. So although the SAM tool is free to use, this last deployment step isn't free and you will need to pay a small fee for S3. Other AWS deployment tools like Cloud Development Kit (CDK) also use the same approach so there is no way to avoid paying for S3 as far as I know. Terraform also creates a similar state file but it either stores it locally or in the Terraform Cloud (with a nice free tier). The latter is especially useful when collaborating in the team. Of course, you can deploy everything manually using AWS interface and it's ok for a quick prototype but if you want to run a CI/CD pipeline, then infrastructure as code is the only option.