kuavi-vqa

name	kuavi-vqa
description	Answer multiple-choice questions about video content using embeddings
argument-hint	<question> -- <choice1> | <choice2> [| choice3] [| choice4]
disable-model-invocation	true

Parse $ARGUMENTS:

• Text before -- is the question
• Text after -- is a pipe-separated list of choices

Example: What color is the car? -- red | blue | green | black

Call kuavi_discriminative_vqa with:

• question: the parsed question text (trimmed)
• candidates: list of parsed choices (each trimmed)

Present the answer with confidence scores for each candidate.