• check the mandatory fields in from the request_body
• validate input of the preprocessing functions

for example:

please assign to me

We can have more validations in few phases like:

1. Validate if connector payload has all the required fields. If not provided, throw the illegal argument exception

2. Then for pre-processing or post processing functions, if the LLM service uri is available url, we need to check if we are using the pre-processing function for the same llm service or not. As an example, for "url": "https://${parameters.endpoint}/openai/deployments/${parameters.deploy-name}/embeddings?api-version=${parameters.api-version}" pre-processing function needs to be: "connector.pre_process.openai.embedding", LLM service is openai here, so we need to use the pre-processing function of openai. Same for post processing function.

3. We can now perform more deeper level check. If we notice in the previous url: the model endpoint is embeddings so we are using the openai.embedding for pre-processing function. Same for post processing function.

We can raise PRs phase by phase accordingly.

A comment from @mingshl :
Just a following up thoughts, if we can check if the credentials is valid in the create API, that would be even better as a further enhancement coming next. Then we can help user ensure the credentials work before calling the model.

@mingshl Can you please elaborate, what kind of validation are you expecting for credentials here?