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

Sinusoidal Positional Encoding.

The original positional encoding from "Attention Is All You Need", using
sine and cosine functions at different frequencies to encode absolute
position information. Deterministic (no learned parameters).

## Formula

    PE(pos, 2i)   = sin(pos / 10000^(2i/d))
    PE(pos, 2i+1) = cos(pos / 10000^(2i/d))

## Usage

    # As an Axon layer that adds PE to input
    encoded = SinusoidalPE.layer(input, dim: 256)

    # Precompute PE table
    pe_table = SinusoidalPE.build_table(max_len: 512, dim: 256)

## References
- "Attention Is All You Need" (Vaswani et al., 2017)
- https://arxiv.org/abs/1706.03762

# `build_table`

```elixir
@spec build_table(keyword()) :: Nx.Tensor.t()
```

Build a sinusoidal positional encoding table.

Returns a tensor of shape [max_len, dim] with precomputed PE values.

## Options
  - `:max_len` - Maximum sequence length (default: 512)
  - `:dim` - Embedding dimension (required)

# `layer`

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

Build an Axon layer that adds sinusoidal positional encoding to the input.

## Options
  - `:dim` - Feature dimension (required)
  - `:name` - Layer name prefix (default: "sinusoidal_pe")

---

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