| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # ai.onnx.Where |
|
|
| `ai.onnx` · standard ONNX operator · ONNX opset ≥ 16 |
|
|
| ## Description |
|
|
| Selects elements from `X` or `Y` according to a boolean `condition`, following NumPy-style multidirectional broadcasting. Where `condition` is `true`, the output takes the value from `X`; otherwise it takes the value from `Y`. |
|
|
| See the [ONNX `Where` spec](https://onnx.ai/onnx/operators/onnx__Where.html) for the reference semantics. |
|
|
| ## Inputs |
|
|
| | Name | Upstream name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | --- | |
| | `condition` | — | `C` | — | — | Boolean mask; nonzero entries select from X, zero entries select from Y. | required | |
| | `x` | `X` | `T` | — | — | Values selected at indices where condition is true. | required | |
| | `y` | `Y` | `T` | — | — | Values selected at indices where condition is false. | required | |
|
|
| ## Outputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `output` | `T` | derived | broadcast result of `condition`, `x`, and `y` | Tensor of shape equal to the broadcasted shape of condition, X, and Y. | required | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `C` | `bool` | |
| | `T` | `float32`, `float16`, `uint32`, `int32`, `int16`, `uint8`, `int8`, `bool`, `int64` | |
|
|
| ## Files |
|
|
| - [`metadata.json`](build/webgpu/metadata.json) — kernel metadata (id, digests, per-variant templates, provenance) |
| - [`manifest.json`](build/webgpu/manifest.json) — the op contract (source of truth) |
| - [`test.json`](build/webgpu/test.json) — correctness cases |
| - [`bench.json`](build/webgpu/bench.json) — benchmark cases |
| - [`where-broadcast.wgsl.jinja`](build/webgpu/where-broadcast.wgsl.jinja) |
| - [`where-vec4.wgsl.jinja`](build/webgpu/where-vec4.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| ```sh |
| npm install --save-exact @huggingface/kernels@0.0.1-preview.3 |
| ``` |
|
|
| Required output shapes and logical data types are inferred from the supplied inputs and attributes; result tensors are allocated automatically. |
|
|
| The `version: 1` option selects the published kernel contract; it is independent of any operator opset, contrib `since_version`, or model version. |
| It follows the `v1` branch as fixes land. To pin exact artifact bytes, pass a 40-character commit `revision` instead of `version`. |
|
|
| Replace each `*Data` placeholder with a typed array containing the corresponding input data. |
|
|
| A `bool` input takes 0/1 values in a `Uint8Array` together with `dtype: "bool"`, since no typed-array class implies it. |
|
|
| ```js |
| import { getKernel } from "@huggingface/kernels"; |
| |
| const kernel = await getKernel("webgpu-kernels/ai.onnx.Where", { version: 1 }); |
| const { output } = await kernel({ |
| condition: { data: conditionData, shape: [3], dtype: "bool" }, |
| x: { data: xData, shape: [1, 3] }, |
| y: { data: yData, shape: [] }, |
| }); |
| ``` |
|
|
