ai.onnx.CausalConvWithState

ai.onnx · standard ONNX operator · ONNX opset ≥ 27

Description

Stateful causal 1-D depthwise convolution. It left-pads each channel with past_state (or zeros), applies a (channels, 1, kernel_size) depthwise convolution and optional SiLU/Swish activation, and returns both the output and the last kernel_size - 1 values as present_state. This inference implementation supports float16 and float32 tensors; bfloat16 is not yet implemented.

See the ONNX CausalConvWithState spec for the reference semantics.

Inputs

Name	Bind key	Logical dtype	Rank	Shape	Description	Presence
`input`	`inputT`	`T`	`3`	—	Input tensor with shape `(batch_size, channels, length)` in channels-first layout.	required
`weight`	`weightT`	`T`	`3`	—	Depthwise convolution kernel with shape `(channels, 1, kernel_size)`, matching the ONNX Conv weight layout for `group = channels`.	required
`bias`	`biasT`	`T`	`1`	—	Optional per-channel bias with shape `(channels)`.	optional
`past_state`	`pastStateT`	`T`	`3`	—	Carry state from the previous step with shape `(batch_size, channels, kernel_size - 1)`. If omitted, the left padding is zero.	optional

Outputs

Name	Bind key	Logical dtype	Rank	Shape	Description	Presence
`output`	`outputT`	`T`	`3`	same as `input`	Convolution output with the same shape as `input`.	required
`present_state`	`presentStateT`	`T`	`3`	derived; see description	Updated carry state with shape `(batch_size, channels, kernel_size - 1)`, including prior state or zero-padding when the current input is shorter than the state.	required

Attributes

Default values (overridable per request):

Attribute	Default	Description
`activation`	`"none"`	Optional fused activation. `silu` and `swish` are aliases; `none` leaves the convolution result unchanged.

Type constraints

Variable	Allowed dtypes
`T`	`float32`, `float16`

Files

metadata.json — kernel metadata (id, digests, provenance)
manifest.json — the op contract (source of truth)
test.json — correctness cases
bench.json — benchmark + tuning cases
causal-conv-with-state-tiled.wgsl.jinja
causal-conv-with-state-vec4.wgsl.jinja
causal-conv-with-state.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.

import { getKernel } from "@huggingface/kernels";

const kernel = await getKernel("webgpu-kernels/ai.onnx.CausalConvWithState", { version: 1 });
const { outputT, presentStateT } = await kernel({
  inputT: { data: inputTData, shape: [1, 1, 2] },
  weightT: { data: weightTData, shape: [1, 1, 5] },
});

Downloads last month: -

kernel

webgpu

wgsl

apache-2.0

WebGPU

Requires WebGPU support. See the compatibility table.