Understand differences in hypotest() test statistic across different backends about pyhf HOT 5 OPEN

def check_runOnePoint_qmu(n_bins):
    pdf, data = generate_pdf_and_data(n_bins)

    q_mu = pyhf.runOnePoint(1.0, data, pdf,
                            pdf.config.suggested_init(),
                            pdf.config.suggested_bounds())[0]

    # Reset the TensorFlow graph and session for each run
    if isinstance(pyhf.tensorlib, tensorflow_backend):
        tf.reset_default_graph()
        pyhf.tensorlib.session = tf.Session()

    return q_mu

def test_backend(backend, n_bins):
    default_backend = pyhf.tensorlib
    pyhf.set_backend(backend)

    test_statistics = [check_runOnePoint_qmu(bins) for bins in n_bins]

    # Reset backend
    pyhf.set_backend(default_backend)

    return test_statistics