How to use the implicit.cuda.CuDenseMatrix function in implicit

def _fit_gpu(self, Ciu_host, Cui_host, show_progress=True):
        """ specialized training on the gpu. copies inputs to/from cuda device """
        if not implicit.cuda.HAS_CUDA:
            raise ValueError("No CUDA extension has been built, can't train on GPU.")

        if self.dtype == np.float64:
            log.warning("Factors of dtype float64 aren't supported with gpu fitting. "
                        "Converting factors to float32")
            self.item_factors = self.item_factors.astype(np.float32)
            self.user_factors = self.user_factors.astype(np.float32)

        Ciu = implicit.cuda.CuCSRMatrix(Ciu_host)
        Cui = implicit.cuda.CuCSRMatrix(Cui_host)
        X = implicit.cuda.CuDenseMatrix(self.user_factors.astype(np.float32))
        Y = implicit.cuda.CuDenseMatrix(self.item_factors.astype(np.float32))

        solver = implicit.cuda.CuLeastSquaresSolver(self.factors)
        log.debug("Running %i ALS iterations", self.iterations)
        with tqdm(total=self.iterations, disable=not show_progress) as progress:
            for iteration in range(self.iterations):
                s = time.time()
                solver.least_squares(Cui, X, Y, self.regularization, self.cg_steps)
                solver.least_squares(Ciu, Y, X, self.regularization, self.cg_steps)
                progress.update(1)

                if self.fit_callback:
                    self.fit_callback(iteration, time.time() - s)

                if self.calculate_training_loss:
                    loss = solver.calculate_loss(Cui, X, Y, self.regularization)

def _fit_gpu(self, Ciu_host, Cui_host, show_progress=True):
        """ specialized training on the gpu. copies inputs to/from cuda device """
        if not implicit.cuda.HAS_CUDA:
            raise ValueError("No CUDA extension has been built, can't train on GPU.")

        if self.dtype == np.float64:
            log.warning("Factors of dtype float64 aren't supported with gpu fitting. "
                        "Converting factors to float32")
            self.item_factors = self.item_factors.astype(np.float32)
            self.user_factors = self.user_factors.astype(np.float32)

        Ciu = implicit.cuda.CuCSRMatrix(Ciu_host)
        Cui = implicit.cuda.CuCSRMatrix(Cui_host)
        X = implicit.cuda.CuDenseMatrix(self.user_factors.astype(np.float32))
        Y = implicit.cuda.CuDenseMatrix(self.item_factors.astype(np.float32))

        solver = implicit.cuda.CuLeastSquaresSolver(self.factors)
        log.debug("Running %i ALS iterations", self.iterations)
        with tqdm(total=self.iterations, disable=not show_progress) as progress:
            for iteration in range(self.iterations):
                s = time.time()
                solver.least_squares(Cui, X, Y, self.regularization, self.cg_steps)
                solver.least_squares(Ciu, Y, X, self.regularization, self.cg_steps)
                progress.update(1)

                if self.fit_callback:
                    self.fit_callback(iteration, time.time() - s)

                if self.calculate_training_loss:
                    loss = solver.calculate_loss(Cui, X, Y, self.regularization)
                    progress.set_postfix({"loss": loss})