single mention · model architecture
Alternating RoPE/NoPE layers
Decoder layers use RoPE in three layers followed by a layer with no explicit positional encoding.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See positional encoding for the methods filed beside it.
Also called 3:1 RoPE to NoPE pattern.
Filed under model architecture › positional encoding