No importer registered for op: Inverse. Attempting to import as plugin, torch.inverse not support to onnx

[lix19937]

import torch
import numpy as np

class TRT_NMS_NN(torch.nn.Module): 
    def __init__(self): 
        super().__init__() 
 
    def forward(self, x, y, z): 
        return torch.inverse(x).matmul(y).matmul(z)    
        # return torch.linalg.inv(x).matmul(y).matmul(z)

sz = 3

x = torch.tensor([[4.1, 2.8, 2.1], [9.676, 6.608, 3.2], [1.1, 2.5, 0.2]])
# x = torch.from_numpy(np.ones(sz*sz).reshape(1,sz,sz)).float()
y = torch.from_numpy(np.arange(sz*sz).reshape(sz,sz)).float()
z = torch.from_numpy(np.arange(sz*sz).reshape(sz,sz)).float()

model = TRT_NMS_NN()
model(x, y, z)

print("export ...")

ops=17
output_file = "inverse" + str(ops) + ".onnx"
torch.onnx.export(
    TRT_NMS_NN(),
    (x, y, z),
    output_file,
    export_params=True,
    keep_initializers_as_inputs=True,
    do_constant_folding=True,
    verbose=True,
    opset_version=ops, #17
)
print("export done")

define torch.inverse as a custom plugin.