LinearParam

Attribute	Type	Default Value	Description
transpose_a	bool	False	-
transpose_b	bool	True	-
has_bias	bool	True	-
out_data_type	torch_atb.AclDataType	torch_atb.AclDataType.ACL_DT_UNDEFINED	The output tensor data type is automatically derived from the input tensors.
en_accum	bool	False	-
matmul_type	torch_atb.LinearParam.MatmulType	torch_atb.LinearParam.MatmulType.MATMUL_UNDEFINED	-

LinearParam.MatmulType

Enumerated items:

MATMUL_UNDEFINED
MATMUL_EIN_SUM

Example

import torch
import torch_atb

def linear():
    m, k, n = 3, 4, 5
    input_tensor = torch.randn(m, k, dtype=torch.float16).npu()
    weight_tensor = torch.randn(k, n, dtype=torch.float16).npu()
    print(input_tensor)
    print(weight_tensor)

    linear_param = torch_atb.LinearParam(has_bias = False, transpose_b = False)
    linear = torch_atb.Operation(linear_param)

    def linear_run():
        linear_outputs = linear.forward([input_tensor, weight_tensor])
        return [linear_outputs[0].to(torch.float32)]


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

if __name__ == "__main__":
    linear()

Parent topic: OpParam