| --- |
| library_name: kernels |
| license: apache-2.0 |
| tags: |
| - kernel |
| - webgpu |
| - wgsl |
| --- |
| # com.microsoft.FusedGemm |
|
|
| `com.microsoft` · ONNX Runtime contrib operator · contrib since_version 1 |
| |
| ## Description |
| |
| Gemm with a fused activation: `Y = act(alpha * A' * B' + beta * C)`, where `A'` and `B'` are optionally transposed and `C` is broadcastable to `(M, N)`. The activation runs in the f32 accumulator before the single output narrowing. This package supports `Relu`, `LeakyRelu`, `Sigmoid`, `Tanh` and `HardSigmoid`; the other activation strings and numeric types admitted by the open schema are not implemented. Omitting `activation` gives plain Gemm. |
| |
| See the [ONNX Runtime `FusedGemm` contrib-operator spec](https://github.com/microsoft/onnxruntime/blob/main/docs/ContribOperators.md#com.microsoft.FusedGemm) for the reference semantics. |
| |
| ## Inputs |
| |
| | Name | Bind key | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | --- | |
| | `A` | `A` | `T` | `2` | — | Left operand, `(M, K)` when `transA` is 0 and `(K, M)` otherwise. | required | |
| | `B` | `B` | `T` | `2` | — | Right operand, `(K, N)` when `transB` is 0 and `(N, K)` otherwise. | required | |
| | `C` | `C` | `T` | — | — | Optional additive term, unidirectionally broadcastable to `(M, N)`: a scalar, a row `(N)`, a column `(M, 1)`, or the full matrix. | optional | |
| |
| ## Outputs |
| |
| | Name | Bind key | Logical dtype | Rank | Shape | Description | Presence | |
| | --- | --- | --- | --- | --- | --- | --- | |
| | `Y` | `Y` | `T` | `2` | derived; see description | `act(alpha * A' * B' + beta * C)` with shape `(M, N)`. | required | |
| |
| ## Attributes |
| |
| Attributes and default values (overridable per request): |
| |
| | Attribute | Default | Description | |
| | --- | --- | --- | |
| | `alpha` | `1` | Scalar multiplier for the product `A' * B'`; the standard default is 1. | |
| | `beta` | `1` | Scalar multiplier for `C`; the standard default is 1. | |
| | `transA` | `0` | Whether `A` is stored transposed. The standard default is 0. | |
| | `transB` | `0` | Whether `B` is stored transposed. The standard default is 0. | |
| | `activation` | — | Optional fused activation name. Supported modes are `Relu`, `LeakyRelu`, `Sigmoid`, `Tanh` and `HardSigmoid`; omission applies none. | |
| | `activation_alpha` | — | First activation parameter: the slope for `LeakyRelu` or `alpha` for `HardSigmoid`. | |
| | `activation_beta` | — | Second activation parameter: `beta` for `HardSigmoid`. | |
|
|
| ## Type constraints |
|
|
| | Variable | Allowed dtypes | |
| | --- | --- | |
| | `T` | `float32`, `float16` | |
|
|
| ## Device requirements |
|
|
| Some implementation variants require `subgroup-matrix` and `subgroups`. These are route-specific capabilities, not package-wide requirements; availability also depends on the request shape and dtype. |
|
|
| ## Files |
|
|
| - [`metadata.json`](build/webgpu/metadata.json) — kernel metadata (id, digests, 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 |
| - [`fused-gemm.wgsl.jinja`](build/webgpu/fused-gemm.wgsl.jinja) |
| - [`gemm-subgroup-matrix.wgsl.jinja`](build/webgpu/gemm-subgroup-matrix.wgsl.jinja) |
|
|
| ## Use with `@huggingface/kernels` |
|
|
| The loader derives every required output's shape and logical dtype from the manifest contract and this call. |
| It then allocates the result tensors automatically. |
|
|
| The `version: 1` option selects the published kernel contract; it is independent of any operator opset, contrib `since_version`, or model 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/com.microsoft.FusedGemm", { version: 1 }); |
| const { Y } = await kernel({ A: { data: AData, shape: [7, 13] }, B: { data: BData, shape: [13, 11] } }, { |
| attrs: { activation: "Relu" }, |
| }); |
| ``` |
|
|
