single mention · model architecture
Three-stage multimodal training protocol
Trains the multimodal model in three stages, introducing multimodal alignment at selected points.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See multimodal architecture for the methods filed beside it.
Also called three stages.
Filed under model architecture › multimodal architecture