RuntimeError: Input type (MPSHalfType) and weight type (torch.HalfTensor) should be the same for RVC with mps about tts-generation-webui HOT 4 OPEN

    if device == "cpu":  # Workaround for "slow_conv2d_cpu" not implemented for 'Half'
        config.is_half = is_half
        config.device = "cpu"

                        ["cuda:0", "cpu", "mps"], label="Device", value="cuda:0"

with gr.Group():
                gr.Markdown("### Hubert")
                with gr.Row():
                    device = gr.Dropdown(
                        ["cuda:0", "cpu", "mps"], label="Device", value="cuda:0"
                    )
                    is_half = gr.Checkbox(
                        label="Use half precision model (Depends on GPU support)",
                        value=False,
                    )
                    gr.Button(
                        value="Clear Hubert (to reload on next generation)",
                        variant="secondary",
                    ).click(
                        fn=lambda: inject_hubert(None)
                    )  # type: ignore

NotImplementedError: The operator 'aten::_weight_norm_interface' is not currently implemented for the MPS device. If you want this op to be added in priority during the prototype phase of this feature, please comment on https://github.com/pytorch/pytorch/issues/77764. As a temporary fix, you can set the environment variable `PYTORCH_ENABLE_MPS_FALLBACK=1` to use the CPU as a fallback for this op. WARNING: this will be slower than running natively on MPS.