Torchscript model predict result is not right

import torch
example = torch.from_numpy(input0_data).to(dtype=torch.float)# torch.rand(1, 10080).cuda()
model = mv_clf.model
script_module = torch.jit.trace(model, example), "")

I use tsai to rain a pytorch model, and save it ,and use the code above to transform the model to torch_script, but when i use triton to inference with the model ,the result is very different from the result using the pkl model. is there any problem with the code saving torch script model??