ElewiseParam

属性	类型	默认值	描述
elewise_type	torch_atb.ElewiseParam.ElewiseType	torch_atb.ElewiseParam.ElewiseType.ELEWISE_UNDEFINED	此默认类型不可用，用户需配置此项参数。
quant_param	torch_atb.ElewiseParam.QuantParam	-	-
muls_param	torch_atb.ElewiseParam. MulsParam	-	-
out_tensor_type	torch_atb.AclDataType	torch_atb.AclDataType.ACL_DT_UNDEFINED	根据输入tensors自动推导输出tensors数据类型。

ElewiseParam.ElewiseType

枚举项：

ELEWISE_UNDEFINED
ELEWISE_CAST
ELEWISE_MULS
ELEWISE_COS
ELEWISE_SIN
ELEWISE_SUB
ELEWISE_EQUAL
ELEWISE_NEG
ELEWISE_QUANT
ELEWISE_LOGICAL_NOT
ELEWISE_ADD
ELEWISE_MUL
ELEWISE_QUANT_PER_CHANNEL
ELEWISE_DEQUANT_PER_CHANNEL
ELEWISE_REALDIV
ELEWISE_LOGICAL_AND
ELEWISE_LOGICAL_OR
ELEWISE_LESS
ELEWISE_GREATER
ELEWISE_DYNAMIC_QUANT
ELEWISE_TANH

ElewiseParam.QuantParam

属性	类型	默认值	描述
input_scale	float	1.0	-
asymmetric	bool	False	-
input_offset	int	0	-

ElewiseParam.MulsParam

属性	类型	默认值	描述
var_attr	float	0.0	-

调用示例

import torch
import torch_atb  

def elewise_add():
    input1_npu = torch.randn(2, 3, dtype=torch.float16).npu()
    input2_npu = torch.randn(2, 3, dtype=torch.float16).npu()
    print("input1: ", input1_npu)
    print("input2: ", input2_npu)
    elewise_param = torch_atb.ElewiseParam(elewise_type = torch_atb.ElewiseParam.ElewiseType.ELEWISE_ADD)
    elewise = torch_atb.Operation(elewise_param)

    def elewise_run():
        elewise_outputs = elewise.forward([input1_npu, input2_npu])
        return elewise_outputs

    outputs = elewise_run()
    print("outputs: ", outputs)

if __name__ == "__main__":
    elewise_add()

父主题： OpParam