| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # ai.onnx.Unsqueeze |
|
|
| `ai.onnx` · standard ONNX operator · ONNX opset ≥ 13 |
|
|
| ## Description |
|
|
| Inserts size-1 dimensions into a tensor at the specified axis positions. The `axes` input lists dimension indices in the output shape where new dimensions of size 1 are inserted; negative values count from the back of the output rank. The output `expanded` has the same data as `data` with rank increased by the number of axes. |
|
|
| See the [ONNX `Unsqueeze` spec](https://onnx.ai/onnx/operators/onnx__Unsqueeze.html) for the reference semantics. |
|
|
| ## Inputs |
|
|
| | Name | Logical dtype | WebGPU storage | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | --- | |
| | `data` | `T` | runtime-selected; narrow integers and bool use 32-bit slots | — | — | Original input tensor whose data is preserved in the output. | required | |
| | `axes` | `S` | `int32` | `1` | — | Logical int64 1-D tensor of axis indices in the output tensor at which to insert size-1 dimensions; negative values count from the back and use int32 WebGPU storage. | required | |
|
|
| ## Outputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `expanded` | `T` | derived | — | Reshaped tensor with the same data as the input and size-1 dimensions inserted at the specified axes. | required | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `T` | `float32`, `float16`, `int32`, `uint32`, `int16`, `uint8`, `int8`, `bool`, `int64` | |
| | `S` | `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 |
| - [`datamove-elementwise-copy.wgsl.jinja`](build/webgpu/datamove-elementwise-copy.wgsl.jinja) |
| - [`datamove-flat-copy.wgsl.jinja`](build/webgpu/datamove-flat-copy.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| ```sh |
| npm install --save-exact @huggingface/kernels@0.0.1-preview.3 |
| ``` |
|
|
| 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: |
|
|
| - `expanded` |
|
|
| 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.Unsqueeze", { version: 1 }); |
| // Explicit destinations request optional results or supply metadata that cannot be inferred. |
| const { expanded } = await kernel({ |
| data: { data: dataData, shape: [] }, |
| axes: { data: axesData, shape: [1] }, |
| }, { |
| outputs: { expanded: { shape: [1], dtype: "float32" } }, |
| }); |
| ``` |
|
|
