single mention · model architecture
NoPE sparse multi-head latent attention
Sparse multi-head latent-attention layers identified as NoPE in a model combining them with KDA linear-attention layers.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See sparse attention for the methods filed beside it.
Also called NoPE sparse MLA layers.
Filed under model architecture › token mixer › sparse attention