deepset-ai/haystack · error · ValueError

The disk device can only be used as a part of device maps

Error message

The disk device can only be used as a part of device maps

What it means

ComponentDevice.from_single raises ValueError when the given Device is a DISK device. A single disk device is meaningless in Haystack; offloading to disk is only expressible as part of a multi-device map (as accelerate uses it).

Source

Thrown at haystack/utils/device.py:286

            The component device representation.
        """
        device = Device.from_str(device_str)
        return cls.from_single(device)

    @classmethod
    def from_single(cls, device: Device) -> "ComponentDevice":
        """
        Create a component device representation from a single device.

        Disks cannot be used as single devices.

        :param device:
            The device.
        :returns:
            The component device representation.
        """
        if device.type == DeviceType.DISK:
            raise ValueError("The disk device can only be used as a part of device maps")

        return cls(_single_device=device)

    @classmethod
    def from_multiple(cls, device_map: DeviceMap) -> "ComponentDevice":
        """
        Create a component device representation from a device map.

        :param device_map:
            The device map.
        :returns:
            The component device representation.
        """
        return cls(_multiple_devices=device_map)

    def _validate(self) -> None:
        """
        Validate the component device representation.

View on GitHub (pinned to e318778c9b)

Solutions

1. Use a real compute device ('cpu', 'cuda', 'mps', 'xpu') for the component
2. If you need multi-device/disk offloading, build a DeviceMap via ComponentDevice.from_multiple/DeviceMap with disk entries
3. Remove 'disk' from single-device config and rely on accelerate's own offloading if applicable

Example fix

// before
comp = ComponentDevice.from_str("disk")
// after
comp = ComponentDevice.from_str("cpu")
# or for disk offloading use a DeviceMap: ComponentDevice.from_multiple(DeviceMap({"layer0": Device(DeviceType.DISK, None), "layer1": Device(DeviceType.CUDA, 0)}))

def component_device_from_str_safe(s: str):
    if s.strip().lower() == "disk":
        raise ValueError("disk is only valid inside a DeviceMap")
    return ComponentDevice.from_str(s)

def is_single_computable_device(d: Device) -> bool:
    return d.type != DeviceType.DISK

try:
    comp = ComponentDevice.from_str(cfg_device)
except ValueError as e:
    print(f"{cfg_device} unusable as a single device: {e}")
    comp = ComponentDevice.from_str("cpu")

When it happens

Trigger: Calling ComponentDevice.from_str('disk') or ComponentDevice.from_single(Device(type=DeviceType.DISK, ...)) — note from_str delegates to from_single.

Common situations: Setting a pipeline component's device to 'disk' in config hoping to offload to disk; translating accelerate device maps that use 'disk' for some layers and mistakenly extracting it as the single component device.

Related errors

• Unknown device type string '{string}'
• Device id must be >= 0, got {id}
• The component device can neither be empty nor contain both a
• Only single devices can be converted to PyTorch format
• Only single devices can be converted to spaCy format

AI-assisted analysis of deepset-ai/haystack@e318778c9b (2026-08-30). Data as JSON: /api/errors/faeaf2fcafbfb981. Report an issue: GitHub.