Enum AttnPositionEncoding 

pub enum AttnPositionEncoding {
    Rope,
    None,
}

2026-09-29: What full attention does to Q and K before the dot product.

Variants

Rope

2026-09-29: Rotary embedding over rotary_dim() dims at rope_theta, or the layer’s RoPE variant (YaRN, proportional, MRoPE) where the loader configures one.

None

2026-09-29: No positional encoding: Q and K enter attention as projected. Nemotron-H attention has none; position reaches it through the Mamba-2 layers only (HF NemotronHAttention.forward never calls apply_rotary_pos_emb).

Trait Implementations

impl Clone for AttnPositionEncoding

fn clone(&self) -> AttnPositionEncoding

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AttnPositionEncoding

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for AttnPositionEncoding

fn eq(&self, other: &AttnPositionEncoding) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Copy for AttnPositionEncoding

impl Eq for AttnPositionEncoding

impl StructuralPartialEq for AttnPositionEncoding