kuavi-vqa

name: kuavi-vqa description: Answer multiple-choice questions about video content using embeddings argument-hint: -- | [| 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.