`concat`¶

Join two or more tensors along one axis.

Category: shape · Identity: concat@1

Shape¶

x* -> out

Relation: out[axis] == sum(x_i[axis]); all other axes equal

Port	Direction	Pattern	dtype
`x`	input	`x*`	compute
`out`	output	`out`	compute

Arguments¶

Name	Type	Default	Constraints	Description
`input_count`	int	required	>= 2	Number of tensors joined; set from the call.
`axis`	int	`1`	>= 0	Axis to concatenate; axis 0 joins along batch and adds the inputs' batches.

Description¶

torch.cat(inputs, dim=axis). Every input is explicit; one missing extent along axis can be inferred from the output contract.

axis=0 joins along batch: the inputs must agree on every other axis and the result carries their batches added together, written 2*B for two equal batches. That is how one shared --- often frozen --- network runs over two branches in a single forward pass; chunk(..., dim=0) takes the result apart again.

Examples¶

Example 1¶

Reorder sections by concatenating them back.

a, b = split(2)
concat(b, a)

Input ['B', 5] → output ['B', 5].

Network: [B, 5] -> [B, 5]  dtype=float32
index  name  operation  input shapes          output shapes
0      n0    split      x=[B, 5]              first=[B, 2], rest=[B, 3]
1      n1    concat     x0=[B, 3], x1=[B, 2]  out=[B, 5]

Parameters: 0

Example 2¶

The second width 6 is inferred.

a = linear(4)
b = linear(x)
concat(a, b)

Input ['B', 8] → output ['B', 10].

Network: [B, 8] -> [B, 10]  dtype=float32
index  name  operation  input shapes          output shapes
0      n0    linear     x=[B, 8]              out=[B, 4]
1      n1    linear     x=[B, 8]              out=[B, 6]
2      n2    concat     x0=[B, 4], x1=[B, 6]  out=[B, 10]

Parameters: 90

Example 3¶

One shared linear runs over both branches at batch 2*B; chunk takes the pair apart.

a = linear(4)
b = linear(x, 4)
pair = concat(a, b, axis=0)
shared = linear(pair, 6)
p, q = chunk(shared, 2, dim=0)
concat(p, q)

Input ['B', 8] → output ['B', 12].

Network: [B, 8] -> [B, 12]  dtype=float32
index  name  operation  input shapes          output shapes
0      n0    linear     x=[B, 8]              out=[B, 4]
1      n1    linear     x=[B, 8]              out=[B, 4]
2      n2    concat     x0=[B, 4], x1=[B, 4]  out=[2*B, 4]
3      n3    linear     x=[2*B, 4]            out=[2*B, 6]
4      n4    chunk      x=[2*B, 6]            out0=[B, 6], out1=[B, 6]
5      n5    concat     x0=[B, 6], x1=[B, 6]  out=[B, 12]

Parameters: 102

concat¶