MATH
·
IN
·
MODELS
The map
Structures
Hypotheses
Classification
Matrix
Methods bubble
Genealogy
Families
Papers
Methods
Why
Cohere
Cohere
Structures found in this family (1)
Linear Subspace
By model (1)
Cohere Embed
Linear Subspace
Observations (1)
Independently-trained embedding models share linear structure (CKA 0.60-0.88)
Papers
Characterizing Linear Alignment Across Language Models (2026)