Open Models Almanac
Methodsmodel architecturemultimodal architecture

single mention · model architecture

Explicit text-string timestamps

Instead of, or beside, a temporal position id, the prompt carries each video patch's time as ordinary text, such as "3.0 seconds". It costs a few tokens per patch, and the model learns to read it like any other token.

Written for the explainer Hearing, speaking and keeping time: audio, omni models and multimodal positions.

Filed under model architecture › multimodal architecture