ConceptAI & ML
GLM-4V architecture adaptation
GLM-4V adapts LLaVA by replacing the projection weight with a stride convolution to support high-resolution inputs while keeping computation in the language model.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
LLaVA1Stride convolution1
Lecturers
1Connections
2The lectures that define it or assume it, and the concepts next to it.