| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # ai.onnx.Slice |
|
|
| `ai.onnx` · standard ONNX operator · ONNX opset ≥ 13 |
|
|
| ## Description |
|
|
| Produces a slice of the input tensor along multiple axes, using `starts`, `ends`, `axes`, and `steps` to select a sub-tensor. Negative indices are resolved relative to the dimension size, and out-of-range values are clamped. Omitting `axes` defaults to all axes in order; omitting `steps` defaults to stride 1. |
|
|
| See the [ONNX `Slice` spec](https://onnx.ai/onnx/operators/onnx__Slice.html) for the reference semantics. |
|
|
| ## Inputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `data` | `T` | — | — | Tensor of data to extract slices from. | required | |
| | `starts` | `S` | `1` | — | 1-D tensor of starting indices for each axis in `axes`. | required | |
| | `ends` | `S` | `1` | — | 1-D tensor of ending indices (exclusive) for each axis in `axes`. | required | |
| | `axes` | `S` | `1` | — | Optional 1-D tensor of axes that `starts` and `ends` apply to; defaults to all axes if omitted. | optional | |
| | `steps` | `S` | `1` | — | Optional 1-D tensor of step sizes per axis; negative steps slice backward, defaults to 1. | optional | |
|
|
| ## Outputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `output` | `T` | same as `data` | — | Sliced data tensor. | required | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `T` | `float32`, `float16`, `uint32`, `int32`, `int16`, `uint8`, `int8`, `bool` | |
| | `S` | `int32` | |
|
|
| ## 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 + tuning cases |
| - [`datamove-slice-block.wgsl.jinja`](build/webgpu/datamove-slice-block.wgsl.jinja) |
| - [`slice-rank2-single-axis-x4.wgsl.jinja`](build/webgpu/slice-rank2-single-axis-x4.wgsl.jinja) |
| - [`slice.wgsl.jinja`](build/webgpu/slice.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| ```sh |
| npm install --save-exact @huggingface/kernels@0.0.1-preview.2 |
| ``` |
|
|
| Outputs with inferable metadata are allocated automatically. Explicit `outputs` entries request optional results or provide metadata that cannot be inferred from the supplied inputs and attributes. |
|
|
| This example supplies explicit metadata for: |
|
|
| - `output` |
|
|
| 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. |
|
|
| ```js |
| import { getKernel } from "@huggingface/kernels"; |
| |
| const kernel = await getKernel("webgpu-kernels/ai.onnx.Slice", { version: 1 }); |
| // Explicit destinations request optional results or supply metadata that cannot be inferred. |
| const { output } = await kernel({ |
| data: { data: dataData, shape: [2, 2] }, |
| starts: { data: startsData, shape: [1] }, |
| ends: { data: endsData, shape: [1] }, |
| }, { |
| outputs: { output: { shape: [1, 2], dtype: "float32" } }, |
| }); |
| ``` |
|
|
