Add support for the KServe inputs format

[ruivieira]

Add support for the KServe "inputs" format.

Example:

Input:

{
    "inputs": [
      {
        "name": "input-0",
        "shape": [1, 4],
        "datatype": "FP32",
        "data": [
          [32, 0, 1, 2]
        ]
      }
    ]
  }

Output:

{"model_name":"example-isvc","id":"04f66a85-a54d-4a9c-aa5e-5f4514cb4def","parameters":{},"outputs":[{"name":"predict","shape":[1,1],"datatype":"INT64","data":[0]}]}

Validation

Assuming

• A KServe installation
• A namespace test
• A test model like the below, deployed in test
• A TrustyAIService CR deployed in test.

Model:

apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
  name: income-isvc
spec:
  predictor:
    model:
      modelFormat:
        name: sklearn
      runtime: kserve-mlserver
      storageUri: "https://github.com/trustyai-explainability/trustyai-explainability/raw/e3a0de0f73bffb764c65e44ccc7f40735104bb9d/explainability-service/demo/models/model.joblib"

And an example payload (e.g. in payload.json):

{
    "inputs": [
      {
        "name": "input-0",
        "shape": [1, 4],
        "datatype": "FP32",
        "data": [
          [32, 0, 1, 2]
        ]
      }
    ]
  }

Issuing an inference request to the model (localhost added for simplicity, adapt to the model's ingress), e.g.

curl -v \
  -H "Host: ${SERVICE_HOSTNAME}" \
  -H "Content-Type: application/json" \
  -d @./payload.json \
  http://localhost:8080/v2/models/income-isvc/infer

Should be correctly parsed by the TrustyAIService. This can be verified either from the TrustyAIService logs, which will report a payload reconciliation, or by inspecting the existence of the data files in the TrustyAIService storage.