[object Object][object Object][object Object]undefined[object Object]

• Description: The DenseLightningIndexerSoftmaxLse operator is a branch operator for the DenseLightningIndexerGradKlLoss operator to compute the Softmax input.

• Formulas:

  $$\text{res}=\text{AttentionMask}\left(\text{ReduceSum}\left(W\odot\text{ReLU}\left(Q_{index}@K_{index}^T\right)\right)\right)$$ $$\text{maxIndex}=\text{max}\left(res\right)$$ $$\text{sumIndex}=\text{ReduceSum}\left(\text{exp}\left(res-maxIndex\right)\right)$$

  maxIndex and sumIndex are transferred to the DenseLightningIndexerGradKlLoss operator as inputs for calculating Softmax.

[object Object]

Each operator has calls. First, [object Object] is called to obtain the input parameters and compute the required workspace size based on the process. Then, [object Object] is called to perform computation.

[object Object]

• Parameters:

  [object Object]

• Returns

  [object Object]: status code. For details, see .

  The first-phase API implements input parameter validation. The following error codes may be returned.

  [object Object]

[object Object]

• Parameters:

  [object Object]

• Returns

  [object Object]: status code. For details, see .

[object Object]

• The data types of the queryIndex and keyIndex parameters must be the same.

• If the weights parameter is not of type float32, the data types of the queryIndex, keyIndex, and weights parameters must be the same.

• Deterministic computation: The default deterministic implementation of aclnnDenseLightningIndexerSoftmaxLse is used.

• Common constraints

  • Processing when the input parameter is empty:
    • If queryIndex is an empty tensor, the function returns directly.
    • The scenarios where the input parameters are empty in the SFAG common restrictions are the same as those in the FAG.
  [object Object]

• Constraints

  [object Object]

• Typ.

  [object Object]

[object Object]

The following example is for reference only. For details, see .