146108a2c027ef2e116236ac8722e00a77103597,thinc/layers/concatenate.py,,forward,#Any#Any#Any#,37

Before Change

def forward(model: Model[InT, OutT], X: InT, is_train: bool) -> Tuple[OutT, Callable]:
    Ys, callbacks = zip(*[lyr(X, is_train=is_train) for lyr in model.layers])
    widths = [Y.shape[1] for Y in Ys]
    output = model.ops.xp.hstack(Ys)

    def backprop(d_output: OutT) -> InT:
        dY = model.ops.xp.ascontiguousarray(d_output[:, : widths[0]])
        dX = callbacks[0](dY)

After Change

def forward(model: Model[InT, OutT], X: InT, is_train: bool) -> Tuple[OutT, Callable]:
    Ys, callbacks = zip(*[lyr(X, is_train=is_train) for lyr in model.layers])
    if isinstance(Ys[0], list):
        return _list_forward(model, X, Ys, callbacks, is_train)
    else:
        return _array_forward(model, X, Ys, callbacks, is_train)


def _array_forward(model: Model[InT, OutT], X, Ys, callbacks, is_train: bool) -> Tuple[OutT, Callable]:
    widths = [Y.shape[1] for Y in Ys]
    output = model.ops.xp.hstack(Ys)

Instances