def forward(self, x):
        caches = []
        h_prev = np.zeros((self.hidden_size, 1))
        h = h_prev

        for t in range(x.shape[0]):
            x_t = x[t].reshape(-1, 1)
            combined = np.vstack((h_prev, x_t))

            z = self.sigmoid(np.dot(self.wz, combined) + self.bz)
            r = self.sigmoid(np.dot(self.wr, combined) + self.br)
            combined_r = np.vstack((r * h_prev, x_t))
            h_ = np.tanh(np.dot(self.wh, combined_r) + self.bh)

            h = (1 - z) * h_prev + z * h_

            cache = (h_prev, z, r, h_, x_t, combined, combined_r, h)
            caches.append(cache)

            h_prev = h

        y = np.dot(self.why, h) + self.by
        return y, caches