ai.onnx.CumSum

ai.onnx · standard ONNX operator · ONNX opset ≥ 14

Description

Computes the cumulative sum of elements along a specified axis. By default, the sum is inclusive (each output element includes itself); setting exclusive to 1 excludes the current element, and setting reverse to 1 scans in the opposite direction.

See the ONNX CumSum spec for the reference semantics.

Inputs

Name	Bind key	Logical dtype	Rank	Shape	Description	Presence
`x`	`x`	`T`	—	—	Input tensor to accumulate.	required

Outputs

Name	Bind key	Logical dtype	Rank	Shape	Description	Presence
`y`	`y`	`T`	same as `x`	same as `x`	Output tensor of the same shape and type as `x`, containing cumulative sums along the specified axis.	required

Runtime arguments

Name	Kind	Semantic	Description	Presence
`axis`	`i32`	`axis`	Axis along which to accumulate; negative values count from the back.	required

Attributes

Default values (overridable per request):

Attribute	Default	Description
`exclusive`	`0`	If 1, the j-th output element is the sum of the first (j-1) input elements, excluding the j-th element itself; defaults to 0 (inclusive).
`reverse`	`0`	If 1, performs the cumulative sum scanning in the reverse direction along the axis; defaults to 0.

Type constraints

Variable	Allowed dtypes
`T`	`float32`, `float16`, `int32`, `uint32`

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
cumsum-splitk-carry.wgsl.jinja
cumsum-splitk-scan-vec4.wgsl.jinja
cumsum-splitk-scan.wgsl.jinja
cumsum-splitk-totals-vec4.wgsl.jinja
cumsum-splitk-totals.wgsl.jinja
cumsum-strided-row.wgsl.jinja
scan-row-chunked-vec4.wgsl.jinja
scan-row-chunked.wgsl.jinja
scan-thread-row.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.CumSum", { version: 1 });
const { y } = await kernel({ x: { data: xData, shape: [5] }, axis: 0 });

Downloads last month: -

kernel

webgpu

wgsl

apache-2.0

WebGPU

Requires WebGPU support. See the compatibility table.