- Description: Retrieves a set of values from weight based on indices, then aggregates them using the specified mode (max, sum, or mean) according to the offsets defined by offsets. Other parameters provide finer control over the computation process.
- Shape derivation formula:
Assume that the shape of the input weight is (numWeight, embeddingDim), the shape of indices is (bagIndices), and the shape of offsets is (bagOffsets).
[object Object]Atlas A3 training products/Atlas A3 inference products[object Object], [object Object]Atlas A2 training products/Atlas A2 inference products[object Object], [object Object]Atlas inference products[object Object], and [object Object]Atlas training products[object Object]:
- When mode is sum:
[object Object]- When mode is mean:
[object Object]- When mode is max:
[object Object]Ascend 950PR/Ascend 950DT:
- When mode is sum:
[object Object]- When mode is mean:
[object Object]- When mode is max:
[object Object]
Each operator has calls. First, aclnnEmbeddingBagGetWorkspaceSize is called to obtain the workspace size required for computation and the executor that contains the operator computation flow. Then, aclnnEmbeddingBag is called to perform computation.
Parameters
[object Object]- For the [object Object]Atlas A2 training products/Atlas A2 inference products[object Object], indices supports only 0D and 1D, and perSampleWeights supports only 1D.
Returns
[object Object]: status code. For details, see .The first-phase API implements input parameter verification. The following errors may be thrown.
[object Object]
- Deterministic computation:
- aclnnEmbeddingBag defaults to a deterministic implementation.
[object Object] and [object Object] support only [object Object].
The following example is for reference only. For details, see .