import torch import pipeline from transformers import pipeline pipe = pipeline( task='onnx-base', model='m3/onnx-base', batch_size=10, device='cuda', ) dummy_input = torch.tensor([[[[1, 2, 3], [4, 5, 6], [7, 8, 9]]]], dtype=torch.float32) input_data = dummy_input.numpy() result = pipe( inputs=input_data, device='cuda', ) print(result)