Enum Operation 

pub enum Operation {
    Input {
        index: usize,
    },
    Constant {
        tensor: WrapDebug<DTensor>,
    },
    View {
        input: Value,
    },
    Broadcast {
        input: Value,
    },
    Permute {
        input: Value,
        permutation: Vec<usize>,
    },
    Slice {
        input: Value,
        axis: usize,
        range: SliceRange,
    },
    Flip {
        input: Value,
        axis: usize,
    },
    Gather {
        input: Value,
        axis: usize,
        indices: Value,
    },
    Concat {
        inputs: Vec<Value>,
        axis: usize,
    },
    Conv {
        input: Value,
        filter: Value,
        details: ConvDetails,
    },
    MatMul {
        left: Value,
        right: Value,
    },
    Unary {
        input: Value,
        op: UnaryOp,
    },
    Binary {
        left: Value,
        right: Value,
        op: BinaryOp,
    },
    Softmax {
        input: Value,
        axis: usize,
    },
    Layernorm {
        input: Value,
        axis: usize,
        eps: Total<f32>,
    },
    Reduce {
        input: Value,
        axes: Vec<usize>,
        op: ReduceOp,
    },
}

The core set of graph operations. Some attempt was made to keep operations orthogonal but flexible, so they can be composed easily.

Variants

Input

A runtime-variable input.

Fields

index: usize

Constant

A constant built into the network.

Fields

tensor: WrapDebug<DTensor>

View

View a value as a different shape.

Fields

input: Value

Broadcast

Repeat along all axes with size 1 that don’t match the output shape.

Fields

input: Value

Permute

Change the order of axis in the shape.

Fields

input: Value

permutation: Vec<usize>

Slice

Slice along the given axis with range start..end.

Fields

input: Value

axis: usize

range: SliceRange

Flip

Flip the given axis.

Fields

input: Value

axis: usize

Gather

Gather values from input at the indices in index on the given axis. indices is a rank-1 tensor.

Fields

input: Value

axis: usize

indices: Value

Concat

Concatenate values along an axis.

Fields

inputs: Vec<Value>

axis: usize

Conv

2D convolution.

Fields

input: Value

filter: Value

details: ConvDetails

MatMul

(Batched) Matrix multiply. If left has shape [b, p, q] and right has shape [b, q, r] the result has shape [b, p, r].

Fields

left: Value

right: Value

Unary

Elementwise unary operation.

Fields

input: Value

op: UnaryOp

Binary

Elementwise binary operation. Both operands must have the same shape.

Fields

left: Value

right: Value

op: BinaryOp

Softmax

Softmax along axis.

Fields

input: Value

axis: usize

Layernorm

Layernorm along axis.

Fields

input: Value

axis: usize

eps: Total<f32>

Reduce

Reduce along the given axes using op. The axes are removed from the shape.

Fields

input: Value

axes: Vec<usize>

op: ReduceOp

Implementations

impl Operation

pub fn inputs(&self) -> Vec<Value>

Trait Implementations

impl Clone for Operation

fn clone(&self) -> Operation

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Operation

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

Formats the value using the given formatter.

impl Eq for Operation

impl Hash for Operation

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 PartialEq for Operation

fn eq(&self, other: &Operation) -> 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 StructuralPartialEq for Operation