# PyTorch GPU
>>> sat_sm = SaT("sat-3l-sm")
>>> sat_sm.half().to("cuda") # optional, see above
>>> text = "this is a test this is another test"
>>> %timeit list(sat_sm.split(text))
# 52.7 ms ± 2.16 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

# PR Patch PyTorch GPU
>>> %timeit list(sat_sm.split(text))
# 2.45 ms ± 24.1 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)

# ONNX CPU
>>> model_ort = SaT("sat-3l-sm", ort_providers=["CUDAExecutionProvider", "CPUExecutionProvider"])
>>> %timeit list(model_ort.split(text))
# 59.3 ms ± 1.83 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)

# PR Patch ONNX CPU
>>> %timeit list(model_ort.split(text))
# 14.7 ms ± 699 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)