# `Edifice.Blocks.DepthwiseConv`
[🔗](https://github.com/blasphemetheus/edifice/blob/main/lib/edifice/blocks/depthwise_conv.ex#L1)

1D depthwise separable convolution block for sequence models.

Depthwise separable convolution factorizes a standard convolution into a
depthwise convolution (per-channel) followed by a pointwise 1x1 convolution.
This reduces parameters from `O(C_in * C_out * K)` to `O(C * K + C * C_out)`.

Used by: Conformer, Mega, StripedHyena, and other hybrid models that need
local pattern extraction alongside attention or SSM layers.

## Architecture

```
Input [batch, seq_len, channels]
      |
Depthwise Conv1D (groups = channels)
      |
Optional BatchNorm / LayerNorm
      |
Activation (SiLU by default)
      |
Pointwise Conv1D (1x1)
      |
Output [batch, seq_len, out_channels]
```

## Usage

    output = DepthwiseConv.layer(input, 256, 31, name: "dw_conv")

# `layer`

```elixir
@spec layer(Axon.t(), pos_integer(), pos_integer(), keyword()) :: Axon.t()
```

Build a depthwise separable 1D convolution Axon layer.

## Parameters

  - `input` - Axon node with shape `[batch, seq_len, channels]`
  - `channels` - Number of input/depthwise channels
  - `kernel_size` - Convolution kernel size (default: 31)

## Options

  - `:out_channels` - Output channels for pointwise conv (default: same as `channels`)
  - `:activation` - Activation function (default: `:silu`)
  - `:use_norm` - Apply layer norm after depthwise conv (default: `true`)
  - `:padding` - Padding mode: `:causal` or `:same` (default: `:causal`)
  - `:name` - Layer name prefix (default: `"depthwise_conv"`)

---

*Consult [api-reference.md](api-reference.md) for complete listing*