core/nn/modalities/multi_modal_classifier library
Multimodal classifier — fuses per-modality features via mean pooling + concatenation + a small MLP head.
Late-fusion recipe (each modality is encoded independently and then combined at the feature level):
AudioTransformer(audio)→ mean-pool →[1, embedA]VideoTransformer(video)→ mean-pool →[1, embedV]- optional
TextTransformer(tokens)→ mean-pool →[1, embedT] concaton axis 1 →[1, embedA + embedV + (embedT?)]Linear(..., numClasses)→ logits[1, numClasses]
The three encoders don't need to share embedDim — the fusion
head sizes itself from their totals.