add transformer tests

    assert isinstance(net(torch.ones(1, 3)), torch.Tensor)

    loss = vi_loss(**pq_to_infer(net))

    assert loss > 0

def test_torch_save_whole_module():

    lin = borch.nn.Linear(2, 1)

    buffer = BytesIO()

    torch.save(lin, buffer)

    buffer.seek(0)

    new_lin = torch.load(buffer)

    assert isinstance(new_lin, borch.Module)

    out = new_lin(torch.ones(2))

    borch.sample(new_lin)

    assert out != new_lin(torch.ones(2))

def test_transformer_samples():

    transformer_model = borch.nn.Transformer(nhead=16, num_encoder_layers=12)

    src = torch.rand((10, 32, 512))

    tgt = torch.rand((20, 32, 512))

    out = transformer_model(src, tgt).sum()

    borch.sample(transformer_model)

    assert out != transformer_model(src, tgt).sum()

def test_transformer_submodules_are_borch_modules():

    transformer_model = borch.nn.Transformer(nhead=16, num_encoder_layers=12)

    for name, mod in transformer_model.named_child_modules():

        import ipdb; ipdb.set_trace()