Interface MultimodalTowers<P>

The towers a multimodal prompt may need; pass only what the request uses (vision-only and audio-only fall out as special cases).

interface MultimodalTowers<P extends { softTokens: number }> {
    audio?: {
        cache?: EncoderCache;
        tokenIds: AudioTokenIds;
        tower: AudioEncoder;
    };
    vision?: {
        cache?: EncoderCache;
        tokenIds: VisionTokenIds;
        tower: VisionEncoder<P>;
    };
}