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

• Function: performs matrix multiplication in the fake-quantization scenario with one input. This API supports only the scenario where the right-hand input matrix of matrix multiplication is in FRACTAL_NZ format.
• Formula:

  • Basic calculation formula:

    $$y = x @ ANTIQUANT(weight) + bias$$

    $weight$ is the input in the fake-quantization scenario, and its dequantization formula $ANTIQUANT(weight)$ is as follows:

    $$ANTIQUANT(weight) = (weight + antiquantOffset) * antiquantScale$$

  • Quantization formula when the output needs to be quantized:

    $$\begin{aligned} y &= QUANT(x @ ANTIQUANT(weight) + bias) \\ &= (x @ ANTIQUANT(weight) + bias) * quantScale + quantOffset \\ \end{aligned}$$

[object Object]

Each operator has calls. First, [object Object] is called to obtain the workspace size required for computation and the executor that contains the operator computation 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]

• Deterministic description: The default deterministic implementation of aclnnWeightQuantBatchMatmulNz is used.
• The following quantization modes are supported: perchannel , pergroup , and mx .
• The input and output support the following data type and shape combinations:

  • Ascend 950PR/Ascend 950DT:

    [object Object]undefined
    • The shape of x is (m, k), the shape of y is (m, n), and the shape of biasOptional is null, (1, n), or (n,).
    • If the data type of weight is INT32 or FLOAT, it indicates tightly packed INT4 or FLOAT4_E2M1, which must meet the following constraints:
      • The last dimension of the original ND matrix is 8-byte aligned.
      • Before calling this API, you must use the [object Object] API to convert the sparsely packed INT32/FLOAT to the tightly packed INT4/FLOAT4_E2M1 and the ND to FRACTAL_NZ. For details, see the example (./aclnnConvertWeightToINT4Pack.md).
      • The shape of the FRACTAL_NZ matrix input to this API is (ceil(n/16), ceil(k/16), 16, 2).
      • Weight supports only non-transposition. The shape of the original ND matrix is (k, n).
    • If the data type of weight is INT4 or FLOAT4_E2M1, it must meet the following constraints:
      • The last dimension of the original ND matrix is 2-byte aligned.
      • Before calling this API, you must use the [object Object] API to convert the ND format to the FRACTAL_NZ format. For details, see the .
      • The shape of the FRACTAL_NZ matrix input to this API is (ceil(n/16), ceil(k/16), 16, 16).
      • Weight supports only non-transposition. The shape of the original ND matrix is (k, n).
    • If the data type of weight is INT8, it must meet the following constraints:
      • The shape of the FRACTAL_NZ matrix input to this API is (ceil(n/32), ceil(k/16), 16, 32).
      • k ranges from 1 to 65535, and n ranges from 2 to 65535.
      • Weight supports both transposition and non-transposition. The shape of the original ND matrix is (n, k) or (k, n).
    • The value of m ranges from 1 to 2^31-1.

[object Object]

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

• If x is of type float16 and weight is of type float32, the [object Object] API needs to be called to assist the calling. The following is an example:

  [object Object]

• The following is an example of calling the [object Object] API to assist in calling when x is of type FLOAT16 and weight is of type INT8.

  [object Object]