| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # ai.onnx.DequantizeLinear |
|
|
| `ai.onnx` · standard ONNX operator · ONNX opset ≥ 25 |
|
|
| ## Description |
|
|
| Dequantizes a quantized tensor back to full precision using the formula `y = (x - x_zero_point) * x_scale`. Scale and zero point determine quantization granularity: scalar for per-tensor, 1-D for per-axis, or same rank as input for blocked quantization. The output type matches `x_scale` unless overridden by `output_dtype`. |
|
|
| See the [ONNX `DequantizeLinear` spec](https://onnx.ai/onnx/operators/onnx__DequantizeLinear.html) for the reference semantics. |
|
|
| ## Inputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `x` | `TQ` | — | — | N-D quantized input tensor to be dequantized. | required | |
| | `x_scale` | `TF` | — | — | Scale for `x`; scalar for per-tensor, 1-D for per-axis, or same-rank tensor for blocked dequantization. | required | |
| | `x_zero_point` | `TQ` | — | — | Zero point for `x` with shape matching `x_scale`; defaults to zero when absent. | optional | |
|
|
| ## Outputs |
|
|
| | Name | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | |
| | `y` | `TF` | same as `x` | same as `x` | N-D full-precision output with the same shape as `x`, typed according to `x_scale` or `output_dtype`. | required | |
|
|
| ## Attributes |
|
|
| Default values (overridable per request): |
|
|
| | Attribute | Default | Description | |
| | --- | --- | --- | |
| | `axis` | `1` | The axis of the dequantizing dimension, used for per-axis and blocked quantization; negative values index from the end. | |
| | `block_size` | `0` | Number of elements along `axis` that share each scale value for blocked quantization; 0 means not blocked. | |
| | `output_dtype` | `0` | ONNX TensorProto element-type code for `y`; `0` uses the data type of `x_scale`. The supported float16/float32 routes require it to agree with `x_scale`. | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `TQ` | `uint8`, `int8`, `int16`, `int32` | |
| | `TF` | `float32`, `float16` | |
|
|
| ## 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 |
| - [`quant-linear-blocked-axis.wgsl.jinja`](build/webgpu/quant-linear-blocked-axis.wgsl.jinja) |
| - [`quant-linear-scalar.wgsl.jinja`](build/webgpu/quant-linear-scalar.wgsl.jinja) |
| - [`quant-linear-vec4.wgsl.jinja`](build/webgpu/quant-linear-vec4.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| ```sh |
| npm install --save-exact @huggingface/kernels@0.0.1-preview.2 |
| ``` |
|
|
| 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. |
|
|
| ```js |
| import { getKernel } from "@huggingface/kernels"; |
| |
| const kernel = await getKernel("webgpu-kernels/ai.onnx.DequantizeLinear", { version: 1 }); |
| const { y } = await kernel({ |
| x: { data: xData, shape: [4] }, |
| x_scale: { data: x_scaleData, shape: [] }, |
| }); |
| ``` |
|
|
