VertexAI Gemini Structured Output with Llamaindex PostProcessor

[hristogg]

Hi,
I am trying to move from genai SDK to VertexAI (what is the point of having two confusing SDKs I am not yet sure).
Specifically for the LLMReranker class which was working fine before with the genai sdk.
However now with the VertexAI class I am not sure how (and whether it is actually possible) to give the generation_config...

Here is what I am trying:
`response_schema = {
"type": "array",
"items": {
"type": "object",
"properties": {
"Document": {
"type": "integer"
},
"Score": {
"type": "integer"
}
},
"required": ["Document", "Score"]
}
}
generation_config=GenerationConfig(
response_mime_type="application/json", response_schema=response_schema
)

rerank_llm = Vertex(
model="gemini-1.5-flash-002",
temperature=0,
max_tokens=8000,
context_window=10000,
additional_kwargs={'generation_config':generation_config} ,
credentials=credentials,
)

ranker = LLMRerank(choice_batch_size=10, top_n=10, llm=rerank_llm, parse_choice_select_answer_fn=custom_parse_chooice_select_fn, choice_select_prompt=rerank_prompt)

`
Which results in the following error:
raise ValueError(
ValueError: Unknown field for GenerationConfig: generation_config

The LLMRerank itself does not seem to have any additional keywords I could use in the postprocess method so I am at a bit of a loss whether this is not yet implemented or I am doing something wrong.