google cloud platform – Vertex AI Batch API input composition

Encountering a weird problem with Vertex AI Batch API from within GCP. Creating a batch job with multiple requests’ JSONL file – the ‘input’ files in GCS look fine, isolated, no issues with provided schemas. Important caveat: it’s a batch with multiple files, each processed via three individual requests with appropriate schemas / prompts. The response from VertexAI arrives as a JSONL GCS file (all jobs SUCCEED), which echoes a super cross-contaminated schema, containing all three schemas in each of the responses. Does it imply that Vertex takes in multiple files but needs a SINGLE prompt + schema combo for each file (gs uri)? Not sure what the issue is – there are no mutability issues in dictionaries on my side. Could not find anything in the docs…