VertexAI: a way to post-process results without usign custom containers?

Google Cloud Build with AI Custom ML & MLOps
1

Hello,

I have been searching on how to deploy models on VertexAI in AI Platform manner. Most tutorials shows using pre-built container which seems to load the model and return inference results.

My current requirement needs to post-process. This was easy to do with AI platform’s Predictor class. Is something like that doable with pre-built containers on VertexAI (where we upload a package for inference by inheriting Predictor class and specifyinh the class name)? Using custom container makes it complex to handle and response to requests.

2

Pre-built containers are convenient to deploy models as Vertex AI provides many pre-built containers but they don’t offer you a chance to customize your workflow beyond the SavedModel. For your scenario, a custom container is the solution as you can build it in your preferred way but it has to comply with several rules required by Vertex AI. [1]

[1] https://blog.ml6.eu/deploy-ml-models-on-vertex-ai-using-custom-containers-c00f57efdc3c