`identity`¶

Pass the tensor through unchanged, as a named node.

Category: activation · Identity: identity@1

Shape¶

x[B, ...] -> out[B, ...]

Port	Direction	Pattern	dtype
`x`	input	`x[B, ...]`	compute
`out`	output	`out[B, ...]`	compute

Arguments¶

This operator takes no scalar arguments.

Description¶

Returns x unchanged: out = x, the same tensor object, with no copy, no parameters, and no effect on the autograd graph beyond passing the gradient straight through.

It exists so a configuration can name a point in the graph — h = identity(name="hidden") gives a branch or a later add something to refer to — and so a slot in a chain can be filled without changing the computation, for instance when a normalization or activation is being ablated. Shape, dtype, and device are preserved exactly for any supported rank: rank 2 [B, F], rank 3 [B, T, D], or rank 4 [B, C, H, W]. Because the shape relation is the identity, it is fully transparent to resolution in both directions. Behavior is identical in train and eval mode.

Examples¶

Example 1¶

A named tap: the passthrough gives a residual add something to refer back to.

h = identity(name="hidden")
linear(8)
relu()
y = linear(4)
add(y, h)

Input ['B', 4] → output ['B', 4].

Network: [B, 4] -> [B, 4]  dtype=float32
index  name    operation  input shapes        output shapes
0      hidden  identity   x=[B, 4]            out=[B, 4]
1      n1      linear     x=[B, 4]            out=[B, 8]
2      n2      relu       x=[B, 8]            out=[B, 8]
3      n3      linear     x=[B, 8]            out=[B, 4]
4      n4      add        a=[B, 4], b=[B, 4]  out=[B, 4]

Parameters: 76

Example 2¶

A no-op placeholder that keeps the shape contract flowing through the chain.

linear(64)
identity()
relu()
linear()

Input ['B', 32] → output ['B', 10].

Network: [B, 32] -> [B, 10]  dtype=float32
index  name  operation  input shapes  output shapes
0      n0    linear     x=[B, 32]     out=[B, 64]
1      n1    identity   x=[B, 64]     out=[B, 64]
2      n2    relu       x=[B, 64]     out=[B, 64]
3      n3    linear     x=[B, 64]     out=[B, 10]

Parameters: 2,762

Example 3¶

Any supported rank passes through unchanged.

conv(8, kernel_size=3, padding=1)
identity()

Input ['B', 3, 8, 8] → output ['B', 8, 8, 8].

Network: [B, 3, 8, 8] -> [B, 8, 8, 8]  dtype=float32
index  name  operation  input shapes    output shapes
0      n0    conv       x=[B, 3, 8, 8]  out=[B, 8, 8, 8]
1      n1    identity   x=[B, 8, 8, 8]  out=[B, 8, 8, 8]

Parameters: 224

identity¶