Enum OpKind 

pub enum OpKind {
    Embed,
    RmsNorm,
    GatedRmsNorm,
    ResidualAdd,
    Copy,
    Concat,
    Split,
    Linear(LinearRole),
    SiluMul,
    Relu2,
    ActQuant(Format),
    QkNorm,
    Rope,
    KvWrite,
    PagedAttention,
    SigmoidGateMul,
    GdnGates,
    Conv1dUpdate,
    L2Norm,
    GdnRecurrence,
    SsmUpdate,
    Router,
    TopK,
    ExpertGateUp,
    ExpertDown,
    Blend,
    EpReduce,
    FinalNorm,
    LmHead,
    Argmax,
    StateSnapshot,
}

2026-09-28: The closed op vocabulary.

Variants

Embed

2026-09-28: Token embedding lookup; a source node.

RmsNorm

2026-09-28: RMSNorm with a learned weight.

GatedRmsNorm

2026-09-28: RMSNorm of one input gated by SiLU of another (GDN output norm).

ResidualAdd

2026-09-28: Elementwise sum of two streams.

Copy

2026-09-28: A copy between buffers.

Concat

2026-09-28: Concatenate inputs along the feature dim.

Split

2026-09-28: Split or deinterleave a packed projection output.

Linear(LinearRole)

2026-09-28: A weight-bearing projection.

SiluMul

2026-09-28: silu(gate) * up over a packed gate|up input.

Relu2

2026-09-29: relu(x)^2 of an ungated up projection.

ActQuant(Format)

2026-09-28: Quantize activations to the given format.

QkNorm

2026-09-28: Per-head RMSNorm of Q or K.

Rope

2026-09-28: Rotary position embedding.

KvWrite

2026-09-28: Append K and V to the paged cache.

PagedAttention

2026-09-28: Paged softmax attention over the cache.

SigmoidGateMul

2026-09-28: x * sigmoid(gate) (attention output gate).

GdnGates

2026-09-28: GDN beta and decay from the ba projection.

Conv1dUpdate

2026-09-28: Causal depthwise conv1d step with its rolling state.

L2Norm

2026-09-28: L2 normalisation of Q and K heads.

GdnRecurrence

2026-09-28: The gated delta rule recurrence over the recurrent state.

SsmUpdate

2026-09-29: The Mamba2 selective state-space update: one step of the SSM state per row (dt softplus, A, the grouped B/C projections and the D skip).

Router

2026-09-28: MoE router logits.

TopK

2026-09-28: Top-k expert selection and weights.

ExpertGateUp

2026-09-28: Routed experts’ gate and up projections.

ExpertDown

2026-09-28: Routed experts’ down projections.

Blend

2026-09-28: Weighted sum of the routed experts and the gated shared expert.

EpReduce

2026-09-28: Cross-rank reduction of an expert-parallel FFN output.

FinalNorm

2026-09-28: The model’s final RMSNorm.

LmHead

2026-09-28: Vocabulary projection.

Argmax

2026-09-28: Greedy token selection over the logits.

StateSnapshot

2026-09-29: A snapshot of the recurrent state its input’s producer updates, taken after each row but the last into the step’s checkpoint slots (the verify’s rollback points). It has no output.

Implementations

impl OpKind

pub fn parse( name: &str, role: Option<&str>, format: Option<Format>, ) -> Result<Self, OpParseError>

2026-09-28: Resolve an op name. linear needs role, act_quant needs format; every other op refuses both.

pub fn base_name(&self) -> &'static str

2026-09-28: The op’s base name (linear, act_quant, rms_norm, …).

pub fn name(&self) -> String

2026-09-28: The canonical spelling with its qualifier, e.g. linear:down, act_quant:fp8/token.

pub fn is_heavy(&self) -> bool

2026-09-28: An op that is never generated: projections, attention, the recurrence, the experts and the vocabulary projection. Fusion happens at its edges.

pub fn reads_linear_weight(&self) -> bool

2026-09-28: Ops that read a weight whose format the precision resolver answers.

Trait Implementations

impl Clone for OpKind

fn clone(&self) -> OpKind

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OpKind

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

Formats the value using the given formatter.

impl Hash for OpKind

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl Ord for OpKind

fn cmp(&self, other: &OpKind) -> Ordering

This method returns an Ordering between self and other.

fn max(self, other: Self) -> Self

where Self: Sized,

Compares and returns the maximum of two values.

fn min(self, other: Self) -> Self

where Self: Sized,

Compares and returns the minimum of two values.

fn clamp(self, min: Self, max: Self) -> Self

where Self: Sized,

Restrict a value to a certain interval.

impl PartialEq for OpKind

fn eq(&self, other: &OpKind) -> 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 PartialOrd for OpKind

fn partial_cmp(&self, other: &OpKind) -> Option<Ordering>

This method returns an ordering between self and other values if one exists.

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

Tests less than (for self and other) and is used by the < operator.

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

Tests less than or equal to (for self and other) and is used by the <= operator.

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

Tests greater than (for self and other) and is used by the > operator.

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

Tests greater than or equal to (for self and other) and is used by the >= operator.

impl Copy for OpKind

impl Eq for OpKind

impl StructuralPartialEq for OpKind